patdavid.net

G'MIC Image Averaging Made (Relatively) Easy

2013-05-10T09:23:00.000-05:00

I hesitate to mention G'MIC in a title sometimes because I feel that whatever I write is not going to do it justice. For this post, I couldn't resist.

I really should have posted this a while ago when I first saw it, but to be honest it slipped under my radar. So I'm trying to catch up and make amends.

+David Tschumperlé was kind enough to post today a link to some incredible work done by Jérome Ferrari with more averaging images. (It is seriously incredible).

A neat result using -compose-edges in G'MIC

This time everything was done entirely in G'MIC. The images are aligned on the eyes automatically using a little cropping and phase correlation, then scaled to fit both eyes. The results are fantastic.

They are also way faster than doing it my way (aligning eyes and scaling to fit by hand in GIMP).

Head on over and check it out:

Averaging face photos : eye alignment

Also, be sure to check the follow up where there are some slight distortions applied to align the mouths as well.

Noise Removal in Photos with Median Stacks (GIMP/G'MIC & Imagemagick)

2013-05-06T15:58:00.000-05:00

In my recent experiments and playing around with even more image averaging in Imagemagick, I decided to have a look at some other methods for calculating pixel values. This time I focused on stacks of images of the same thing.

Why the same thing?

f/8 @ ¹⁄₁₀₀ sec., ISO 25,600 (mouseover to compare straight out of the camera)

Because the (uninspired) above image was shot at ISO 25,600. Go look at it again.

To be fair, I cheated a little bit... The relatively noise-free image you see above shot at ISO 25,600 is actually combined from 10 identical shots.

The trick is that I combined all of the shots using a median evaluation in Imagemagick. This basically means that for every pixel in the image, there were 10 pixel values to choose from. Imagemagick automatically picked the pixel value that was closest to the middle.

So if we had a list of 7 values, for example, that looked like this:

120, 120, 125, 137, 150, 151, 160

Then the median value is just the number in the middle of the list:

120, 120, 125, 137, 150, 151, 160

If there were an even number of values, the median is calculated as the average of the two numbers in the middle (usually).

This is different from the averaging I had done previously, where the pixel value was an average of all the values at a given pixel location (though both methods could be used to reduce noise in the final result).

Here, let's look at a couple of 100% crops side by side between my median stack example and a base image:

Pretty neat, right? (Seriously, look at that last image - you can't even make out "Fast Focused Fearless" in the single shot)

There's a few obvious caveats, though. This is only really effective for static scenes (we'll see why shortly), and the images must be aligned very well for it to work without smearing any fine details. So it should be shot from a tripod or very stable place to minimize movement between shots, or the shots should be aligned in post (using Hugin's align_image_stack - see here for an example of how to do it).

How to do it

There are two ways you can go to achieve this. The fast way, and the slow way.

First make sure your images are well aligned - shoot on a tripod if at all possible, and/or align the images with software as needed. Take as many as you feel like waiting on.

The fast way is with Imagemagick. Once you've aligned all your images (or they are spot on already), put them into a directory. Then in that directory:

convert *.jpg -evaluate-sequence median OUT.jpg

The slow way is using G'MIC in GIMP. Load up each of your aligned images as a layer in GIMP, then run G'MIC.

Then go to Layers → Blend [median] and make sure the Input Layers are All visibles, and optionally output to New layer(s).

Be prepared to wait, I've found the G'MIC version of this blend is 10x+ longer than doing it through Imagemagick. (Though I did poke David about this - so the G'MIC team might optimize this blend mode later).

Removing People

Using a Median filter to combine multiple images has been used to remove spurious objects from a scene. A common example is if you desire to get a shot of a famous landmark or tourist area, where there tends to be a lot of people around.

You can just shoot multiple shots (try for an odd number), and combine them with a median filter to remove the spurious objects throughout the scene.

For instance, here's my well-used space marine traversing my desktop:

As you can see, the final frame is after combining all of the images with a median evaluation. Voila! No more space marine!

In Conclusion

This is a really neat method to use, once you realize it's available to you, and where it can be best utilized.

If you're shooting static scenes (stock photography, still life, etc.), there's really no reason not to go ahead and set your camera to fire off a handful of frames quickly (burst mode!). Setup on a tripod, burst away, and you can significantly reduce the noise in your images with little effort.

If you want to capture scenes with many people/objects moving about, and want a clean image of an environment, just setup on a tripod again, and snap away. With enough frames, it'll make combining the images and removing those objects very easy to do.

[Addendum]
Just a couple of quick things that a few people have pointed out:

1. Can't I just use a lower ISO and longer exposure?
Yes - but what happens when you've pushed your ISO down as far as it can go on your camera? With this method, you can remove even more noise and make images that are cleaner than is possible with a single shot with your camera.

2. This is not far off from what astrophotographers will also do to clean up ^signal⁄_noise in their images - stack multiple images to clean up the noise. In fact - there is specialized software just for this type of image stacking that I should go try out! :)

Bouncing Baby Beverly Tutorial (Making Babies Fly)

2013-04-19T10:37:00.000-05:00

Way back in November, 2011 I had written a tutorial for PetaPixel that walked through how I created the image of my daughter bouncing out of her crib:

Bouncing Baby Beverly by Pat David, on Flickr

I realized that the tutorial only existed over on PetaPixel, and that I should probably just put a copy of it here for anyone interested. So here it is.

The rest of my GIMP tutorials can be found here:
Getting Around in GIMP
My original post of Bev with more examples can be found here.

An earlier post here on PetaPixel showcased the wonderful image created by Indylab of a flock of cell phones and the method used to create it. In a rather snarky comment, I said to get back to me when they started tossing babies, and linked to my daughter merrily jumping in her crib with her toys. Mike was kind enough to approach me about writing up a small walk-through on how I created my image, and who can honestly turn down a chance to show off their baby daughter looking so cute?

Conception

I had this image in my head for quite some time before I actually got a chance to shoot it. It was the result of two things mainly:

A desire to practice post processing images for compositing.
A desire to buy a tripod.

My wife let me buy a tripod (for other long-exposure experiments including light painting), but only because I justified it with “a neat idea” for a picture with our daughter (the “I have a great idea for a photo of our daughter, but it will require me to buy more camera gear” defense).

My desire to attempt this shot was born out of how cute I think my daughter is, and her personality is almost always explosively positive and joyful. So in my head I was imagining her just bursting with happiness so much that she literally would be jumping right off her crib. It only logically follows that if she was jumping out of that crib, her stuffed animals would be as well!

Execution

I only had a rudimentary vision of what I wanted the final image to look like, but it was enough to nail down the overall idea. In thinking about framing up the image, I liked the idea of her crib being squared up straight on in the image. I wanted the focus to be on her and her animals, and didn’t want anything else detracting from that.

I set up my tripod and camera (Oly E-P1 w/ 17mm f/2.8) in her room and framed up my shots. I had a DIY softbox for my flash precariously balanced on the bureau off screen to the right (wirelessly triggered), and a large window directly behind the camera lighting the scene.

To start things off I needed a single image of the scene without any of the flying objects in it. This image would be the base I would use to later mask out anything I didn’t want (namely, me holding her). I used a spare outtake shot of one of the stuffed animals that was nowhere near my scene as my base.

Next set of images were of me holding her over her crib in various positions. Luckily she’s a really good sport about it, and a lot of tickling and kisses between takes didn’t hurt (hint: most models do not take kindly to this, use caution).

Multiple takes were done in order to have the highest chance of a good image of her, and to give me multiple textures to use in post later to assist in masking me out of the image. The trick is to pre-visualize what level of work will be required in post, and to accommodate that when shooting.

For instance, I knew it would be a pain to clean up the edges around her clothing where I was holding her later, so I tried to minimize my hands and arms obscuring her too much.

The rest of the images were quite easy - throw stuffed animal into frame, and fire! Rinse, repeat as necessary (again, multiple shots really pays off here - each animal probably had about 8 or so images made).

The trick to making life easier in post is to make sure that the lighting and exposure is controlled to be as consistent as possible between shots - you really want everything lit with a consistent power and direction.

Post Processing

This is really where the vision begins to come together for me. I’ll usually fire up my image editor (GIMP in my case, PS for those of you with a little extra cash), and set my first image of just the background environment as my base layer.

This first image with nothing on it becomes the background against which I will begin masking out objects I do not want from subsequent layers. I loaded up each of my other shots as layers in the image to get a rough idea of placement within the scene (some stuffed animals would overlap each other in some shots - I picked through all the images until I found ones that I liked).

Then I start working on removing myself from the image. With an environment background this is relatively quick to add a layer mask in GIMP and paint over myself on the image to show the base image through in a big, rough way (the only place I slowed down was nearing the edges).

The devil is in the details, they say, and nowhere is this more apparent than when trying to make a clean edge mask from a background.

There really is no quick and easy shortcut to getting good results here - a small radius brush and patience is your friend. Too hard a brush and the edge looks fake immediately, too soft a touch and the edge looks fuzzy and fake immediately. Somewhere in between is a happy medium, and that’s what I was aiming for.

Luckily, I remembered to take multiple shots, so in places where the edges were not cleaning up as well as I would have liked I “borrowed” portions of the other shots to make the edge cleaner (the side of her body by the bib is from another shot, as is her sleeve). When working on edges like this it really does pay off to have multiple takes to draw from!

Once I had blocked myself out of the image, I continued the process by adding layer masks to each of the stuffed animals, and painting them back into the visible image (by modifying the layer masks). No overlapping animals made this a bit easier, but you’ll notice that something key is missing from this version of the image.

Shadows!

That whole devil and details thing again. The trick now was only to create some fake shadows for Beverly and her animals that didn’t have shadows (the doll at top left already had one).

The shadows are key in lending a sense of realism to the final product. They were made by simply approximating the shape of the shadow, filling with an appropriate shadow color (sampled from other shadows on the back wall), and blurring to taste. After that I just adjusted opacity by eye until it all seemed reasonably well blended.

Overall I was happy with how this image turned out, and it really was fun creating it (despite what looks like a lot of tedious work). I’ve since done others along a similar vein, but this first one still makes me smile when I look at it...

Using Imagemagick to Create Contact Sheets (Montage)

2013-04-09T12:12:00.000-05:00

In my previous post on postprocessing my Open Source portrait, I had shown a couple of contact sheets that I had generated to help me narrow down which images I liked. That contact sheet looked like this:

Contact sheets are a great holdover from the days of film, when you wanted a big overview of all the shots on a strip of film. After developing the film, you could just lay a sheet of photo paper under the enlarger, lay your strips of film on top, put a glass plate over the whole thing to hold them down, and then expose with white light for a small amount of time.

This would produce a single sheet with all your images as thumbnails:

David Hume Kennerly [Public domain or Public domain], via Wikimedia Commons

This is probably redundant these days with the ability to look at a directory listing as thumbnail images, but I still find uses for this type of image.

My personal photos are kept on a backup server. That computer has no monitor, keyboard, or mouse. The main way I access it is through shares on my network. If I'm off-site, I'll access my network using SSH.

The upside is that it's wayyy faster than using a GUI. The downside is that there's no GUI...

The machine I'm SSH'ing into does have my Dropbox account linked, though. So what I do is run Imagemagick Montage on the machine I remote into, and have it generate a "contact sheet" for me, placing it into my Dropbox folder.

Wait a few minutes, and voila! Instant contact sheet for me to peruse on my local machine. Here is what my montage command looks like:

montage -verbose -label '%f' -font Helvetica -pointsize 10 -background '#000000' -fill 'gray' -define jpeg:size=200x200 -geometry 200x200+2+2 -auto-orient P21603{65..70}.JPG ~/Dropbox/contact-dark.jpg

Yes, it looks scary. I promise you, it's not. As usual, let's have a look at the command line option by option to see whats going on...

montage -verbose -label '%f' -font Helvetica -pointsize 10

Fire up the montage command. Set it to -verbose output to see whats going on.

Label all the thumbnails with the filename, using the font Helvetica at a point size of 10. See, not too bad so far!

-background '#000000' -fill 'gray'

Set the background color to black (#000), and the fill color for the label text to gray.

-define jpeg:size=200x200

This one is interesting. I found that if I tried to run the command on a full directory of jpegs, I would quickly run out of memory. This is because imagemagick will read in each file and keep the full size image in memory until it's done. Holy Mackerel!

So this option tells imagemagick to resize the jpeg as soon as it's read in, and to only keep the smaller version in memory. This significantly speeds up the process of generating the contact sheet.

-geometry 200x200+2+2

This just tells imagemagick to set each thumbnail size on the final output to be 200x200 pixels, with a 2px border around them.

-auto-orient

Rotate any images to their correct orientation as specified in their metadata.

P21603{65..70}.JPG

Now this is interesting. I'm not sure if this will work in Windows, but it does on a BSD/OSX/*nix shell. I can specify a numerical range of images that I want to be in my final output.

The above command will basically be expanded so that the input files are:

P2160365.JPG
P2160366.JPG
P2160367.JPG
P2160368.JPG
P2160369.JPG
P2160370.JPG

If you wanted to just do all the images in the directory, you could use the normal glob syntax for all JPG images:

*.JPG

I did this to limit the final size of the output contact sheet to a range of images. For both final size as well as memory limits on my machine (especially if I had hundreds of images in my directory.

I could also have done it by using the glob syntax. For instance, if I wanted all the images in the 600-699 range, I could have used:

P21606*.JPG

Or any variation on that (I'm not going to get into the syntax here, but there are plenty of references around for it).

~/Dropbox/contact-dark.jpg

Output the final image "contact-dark.jpg" to my Dropbox folder.

That's about it. That command puts out a nice contact sheet for me to quickly eyeball and find images that I like. From there I can narrow it down and look more closely at the final picks.

I could even go so far as to use this command as part of a shell script that can recursively walk through all of the folders of my photos, and to output contact sheets for each one if I wanted. I may get into that at another time...

Getting Around in GIMP - Blue Channel Check Layer (Retouching)

2013-04-01T17:24:00.001-05:00

In my previous post for the Open Source Portrait (Postprocessing), I had mentioned during the skin retouching in GIMP section that there was a method that I had used in the past for helping to make skin imperfections visible for healing.

The rest of my GIMP tutorials can be found here:
Getting Around in GIMP

It is a similar method described by Calvin Hollywood (again), that he calls "Check Layers". He explains what they are and how he uses them in the video below:

If you don't want to watch the video, the basic idea came about from my experience when converting images to B&W. That is, the Blue channel of a color image is not flattering at all for skin (unless you have flawless skin). Conversely, the Red channel usually has very pretty skin tones (and hence it's common to overlay a Red channel on your image and set it to Overlay blend mode to enhance this).

What Calvin Hollywood shows in his video is using a Photoshop adjustment layer to turn the image into grayscale, and to turn down the contribution from the Red channel, while increasing the Blue channel. This results in a more contrasty image that favors the blue channels, and will make skin imperfections much more visible.

The problem is that we don't have adjustment layers in GIMP...

The Solution

Well, I had to play around a bit to get things working, but I finally got there.

The trick was to remember that I wanted a view of the Blue channel, in monochrome, and in a way that any touchups I did to the base image would be instantly reflected in my grayscale Blue channel view.

Here's how to do it (manually, I have also written a script-fu you can download to do this automatically - see below):

Start with your base layer (or a duplicate).
Add a new layer over it, fill it with pure Red (255,0,0), layer mode set to Subtract.
Add another new layer, fill it with pure Green (0,255,0), layer mode set to Subtract.
Add a new layer over it, fill it with pure Yellow (255,255,0), set layer mode to Subtract.

Reader Sebastian pointed out in the comments the (now) obvious fact that Yellow is the complement of blue, so this saves having to make 2 layers of red and green - we can just subtract the complement of blue: Yellow.

Add another new layer, fill it with White (255,255,255), layer mode set to Color.
Add another new layer, fill it with Grey (127,127,127), layer mode set to Dodge.
Now work with your healing brush on your base layer, and changes will show in realtime.

When I'm done adding layers, this is what my layer palette would look like:

After adding all my layers

I can now work with a healing/clone brush on my base layer, and I'll see the changes reflected in real time on my canvas in GIMP.

If I want to see the progress of my work on the full color image, I can use a simple shortcut: Shift-Click (⌘-Click) on the base layer visibility to isolate it as the only visible layer, and again to turn all layers back on.

Here is a quick video of me walking through using this method:

A nice bonus to this method is that it will also work above a set of Wavelet Decompose scales, meaning you can use a normal wavelet workflow, and still use this "Check Layer" to get a view of what you're doing against the Blue channel.

Finally, it can be tedious to create all of these layers manually each time you want to use it, so I've written a Script-Fu to automate it. You can find it at registry.gimp.org:

Download the Script-Fu from the GIMP Registry

The Open Source Portrait (Postprocessing)

2013-03-28T14:42:00.001-05:00

In my previous post I walked through the concept and shoot of my portrait of Mairi. At this point all I've got are a bunch of RAW (and jpg) files of the shoot. I know that I forgot to mention it in the last post, but for goodness sake if you can shoot RAW - do it. (It saved my butt with this image, as you'll see below).

Also, you'll be able to download my RAW file, and JPG output from RawTherapee below.

The previous tutorial in this series is here:
The Open Source Portrait (Equipment & Environment)

This is a long post.
Seriously, this is the longest post I've ever written.
I'm walking through many of the things I had previously written tutorials for, and apply them to a process this time so you can see them as part of a workflow.
So, here's a Table of Contents so you can jump to the section you need:

Picking Your Image
RAW Processing

Adjust Exposure

Exposure Compensation
Black Point

White Balance
Noise Reduction
In Summary

GIMP Retouching

Touchup Hair
Fixing the Background/Cropping
Skin Retouching & Wavelet Decompose

In Summary

Contour Painting Highlights
Color Curves
Sharpening

The End

File downloads:

Download the .ORF RAW file [Google Drive]
Download the full resolution JPG from RawTherapee.
Download the Full Resolution .XCF file [.7zip - 265MB]
If you want to use the .XCF file just to see what I did, I recommend the ½ resolution file, as it's smaller.
Download the ½ Resolution .XCF file [.7zip - 60MB]

These files are being made available under a Creative Commons Attribution, Non-Commercial, Share Alike license (CC-BY-SA-NC).
You're free to use them, modify them, and share them as long as you attribute me, Pat David, as the originator of the file. You're not allowed to use these images for Commercial purposes.

And to whet your appetite, here is the final result of all of the postprocessing of my portrait (mouseover to compare it to no retouching):

The final result I'm aiming for
Mouseover to compare to unretouched version

Picking Your Image

This is a hard thing to quantify, as each of us is driven by our own vision and style. In my case, I wanted something a little more somber looking with a focus on her eyes (they are the window to the soul, right?). There's just something I like about big, bright eyes in a portrait, particularly in women.

I also personally liked the grey sweater against the grey background as well. I felt that it put more focus on the colors of her skin, hair, and eyes. So that pretty much narrowed me down to this contact sheet:

Narrowing it down to this set.

Looking over the shots, I decided I liked the images with the hood up, but her hair down and flowing around her. This puts me in the top two rows, with only a few left to decide upon. At this point I narrowed it down to one that I liked best - grey sweater, hood up but not pulled back against her head, hair flowing out of it, and big eyes.

This is pretty common, I'd imagine. You can grab several frames, but in the end hopefully just the right amount of small details will come together and you'll find something that you really like. In my case it was this one:

I finally decided on this shot based on the color, hair, eyes, and slight smile.

Now hold on a minute. The image above is the JPG straight out of the camera. As you can see, I've underexposed this one a little bit, and the colors are not anywhere near where I'd like them to be. If you're following along don't download this version of the image. I'll have a much better starting JPG after we run it through some RAW development first!

If you're impatient, jump to that section and get the image there.

Raw Processing

There are a few RAW conversion options out there in the land of F/OSS. Here's a small list of popular ones to peruse:

One of the reasons I love using F/OSS is the availability (usually) of the software across my OS's. In my case I went with RawTherapee a while back and liked it, so I've stuck with it so far (even though I had to build my own OSX versions).

So, my workflow includes RawTherapee at this point. You should be able to follow along in other converters, but I'm going to focus on RT because that's what I'm using. If you shoot only in JPG (seriously, use RAW if you can), you can skip this section and head directly down to GIMP Retouching.

Load it up

After starting up RawTherapee, you'll be in the File Browser interface, waiting for you to select a folder of images. You can navigate to your folder of images through the file browser on the left side of the window. It may take a bit while RawTherapee generates thumbnails of all the images in your directory.

RawTherapee file browser view (Navigate folders on the left pane)

Once you've located your image, double clicking it in the main window will open it up for editing. If you're using a default install/options on RT, chances are a "Default" profile will be applied to your image that has Auto Levels turned on.

The base image with "Default" profile applied (auto levels)

Chances are that Auto Levels will not look very good. My Default processing profile usually does not look so hot (no noise reduction, auto levels, etc.). That's ok, because we are going to fix this right up in the next few sections.

Adjust Exposure

I like to control the exposure and processing on my RAW images. Auto Levels may work for some, but once you get used to some basic corrections and how to use them it's relatively quick and painless to dial-in something you like quickly.

Again - much of what I'm going to describe is subjective, and will depend on personal taste and vision. This just happens to be how I work, adjust as needed for you own workflow. :)

To give me a good starting point I will usually remove all adjustments to the image, and reset everything back to zero. This is easy to do as my Default profile has nothing done to it other than Auto Levels.

Auto Levels values on the Exposure panel

A quick and easy way to reset the Exposure values on the Exposure panel is to use the Neutral button on that panel (I've outlined it in green above).

You can also hit the small "undo" arrows next to each slider to set that slider back to zero as well (for some reason the Neutral button doesn't want to reset the Saturation slider to zero, so I'll do that manually.

At this point the image exposure is set to a baseline we can begin working on. For reference, here is my image after zeroing out all of the exposure sliders and the saturation:

With all exposure adjustments (and saturation) set to zero

Exposure Compensation

The first thing I'll begin adjusting is the Exposure Compensation for the image. You want to be paying careful attention to the histogram for the image to know what your adjustments to Exposure Compensation are doing, and to keep from blowing things out.

I personally begin pushing the Exposure Compensation until one of the RGB channels just begins butting up against the right side of the histogram. Here is what the histogram looks like for the neutral exposure:

Neutral exposure histogram

After adjusting Exposure Compensation I get the Red channel snug up against the right side of the histogram:

Exposure Compensation until the values just touch the right side

If you go a little too far, you'll notice one of the channels will spike against the side, and if you really go too far, you'll get a small colored box in the upper right corner indicating that channel has gone out of range (is blown out).

So here is what my image looks like now with only the Exposure Compensation adjusted to a better range:

Exposure Compensation adjusted to 2.40

The Exposure panel in RT now looks like this (only the Exposure Compensation has been adjusted):

Exposure Compensation set to 2.40 for this image

If the highlights in your image begin to get slightly out of range, you may need to make adjustments to the Highlight recovery amount/threshold, but in my case the image was slightly under-exposed, so I kept it zero.

There is also a great visual method of seeing where your exposures for each channel are at, and to avoid hightlight/shadow clipping. Along the top of your main image window, to the right, there are some icons that look like this:

Channel previews, Highlight & Shadow clipping indicators

The Channel previews let's you individually toggle each of the R,G,B, and Luminosity previews for the image. You can use these with the Highlight and Shadow clipping indicators to see which channels are clipping and where.

Highlight and Shadow clipping indicators will visually show you on your image where the values go beyond the threshold for each. For highlights, it's any values that are greater than 253, and for shadows it's any values that are lower than 8.

To illustrate, here is what my image looks like in RT with the Exposure Compensation set to 2.40 from above:

With Highlight & Shadow clipping turned on

I don't mind the shadows clipping in the dark regions of the image, though I can make adjustments to the Black Point (below) to modify that. The highlight clipping on her face is of more concern to me. I certainly don't want that!

At this point I can dial in my Exposure Compensation for the highlights by backing it down slightly. As I ease off it I should be seeing the dark patch for Highlight Clipping growing smaller. I'll stop when it's either all gone, or just about all gone.

I wasn't too far off in my initial adjustment, and only had to back the Exposure Compensation off to 2.30 to remove most of the highlight clipping.

Settings so far (everything else zero)...

Exposure Compensation

2.30

Black Point

At this point I will usually zoom a bit into a shadow area of my image that might include dark/black tones. The blacks feel a little flat to me, and I'm going to increase the black level just a bit to darken them up.

I want to be zoomed in a bit so I can determine at which point the black point crushes any details that I want to be visible still. You want your blacks to be dark if possible, but you want to keep details in the shadows if possible (it's really, really subjective where this point is, but I'll err on the conservative side since I am still going to process colors a little bit in GIMP later).

Starting with a Black point of zero:

I will increase the Black point while keeping an eye on those shadow details, increasing it until I like how the blacks look and I haven't destroyed detail in the dark tones. I finally settled on a Black value of 150 as seen here:

Black value set at 150 (still keeping sweater details in the shadows).
Mouseover to compare to previous.

Watch out for Shadow Recovery when you first start adjusting the Black Point. It's default might be a different value than zero (mine is at 50), and the Neutral button won't set it back to zero (resetting it will give it back to it's default value of 50). You may want to push it manually to zero, and if you feel you want to bump shadow details a bit, then start pushing it up.

I know things look noisy at the moment, but we'll deal with that in the next section (there is no noise reduction being applied at this point).

Settings so far (everything else zero)...

Exposure Compensation	2.30
Black	150

Brightness, Contrast, and Saturation

For this image I didn't feel the need to modify these values, but this is purely subjective (again). If you do modify these values, keep an eye on the histogram and what it's doing to keep things from getting out of range/whack again.

White Balance

Hopefully you had the right White Balance set during your shoot in camera. If not, it's ok - we're shooting in RAW so we can just set it as needed now.

I happen to have had my in-camera WB set to Flash, so the embedded WB settings in my RAW file metadata are pretty close. In my shot, however, you'll notice that there is a bit of a white window visible in the left of the frame. I happen to know that the window is quite white, and should be rendered as such in my image.

As a side note, what I really should have done was to get myself a good reference for balancing the white balance, and to shoot it as part of my setup. Something like the X-Rite MSCCC ColorChecker Classic, or even a WhiBal G7 Certified Neutral White Balance Card. These are a little pricey, but any good 18% grey card will do, really. I just happen to know that my window borders are a pure white, so I'm cheating a bit here...

So here is what our image looks like at the moment:

Image so far, with White Balance set to Camera (Default)

The White Balance for your image can be adjusted from the Color panel:

Default Color panel showing Camera white balance

You can try out some of the presets in the Method drop-down - there are the typical settings there for Sunny, Shade, Flashes, etc... In my case I am going to use the Spot WB option. Clicking that button will let me pick a section of my image that should be color neutral.

In my case, I know that the window border should be white (and color neutral), so I will pick from that area on my image. Doing so will shift my WB, and will produce a result that looks like this:

WB based on white window border (mouseover to compare Camera based)

I also happen to know that the grey colored walls in the background are close to neutral, but with the slightest hint of blue in them. If I used the grey wall instead of the white window, I would introduce the slightest warm cast to the image. I tried it (choosing a section of the grey wall on the right side of the background), and actually prefer the slightly warmer color, personally:

WB based on the grey wall background (right side of image).
Mouseover to compare to window WB

The difference is ever so slight, but it is there. In my original final image, I went with the balance pulled from the wall, so I will continue with that version here. If you're curious, here is what my WB values look like:

After setting Spot WB to the window

Seriously, though, don't rely on luck. Get a grey/color card to correct color casts if you can...

Settings so far (everything else zero)...

Exposure Compensation	2.30
Black	150
WB Temperature	7300
WB Tint	0.545

Noise Reduction & Sharpening

Chances are the RAW image is going to look pretty noisy zoomed in a bit. This isn't unusual since we are dealing with RAW data. There are two noise reduction (NR) options in RT, and we are going to want to use both.

Impulse Noise Reduction

This NR will remove pixels that have a high impulse deviation from surrounding pixels. Basically the "salt and pepper" noise you may notice in your images where individual pixels are oddly brighter/darker than the surrounding pixels.

If I zoom into a portion of my image (not far from where I was looking at shadows for setting a black point), I'll see this:

Closeup crop with no Impulse Noise Reduction

I'll normally play a bit with the Impulse NR to alleviate the specks while still retaining details. As with most NR methods - going a bit too far will obliterate some details with the noise. The trick is to find a happy medium between the two. In my case, I settled on a value of 55 (the default is 50):

Impulse NR set to a value of 55 (mouseover to compare)

I could have gone a bit further (and have in others from this series), and pushed it up to the 60-70 range, but it's a matter of taste and weighing the tradeoffs.

Luminance/Chrominance Noise Reduction

These two NR methods will suppress noise in the luminance channel (brightness), and the blue/red chrominances.

I will use a light hand with these NR values. The defaults are 5 for each, and it should make a noticeable difference just with the default values. If you push the Luminance NR too far, you'll smear fine details right off your image. If you push the Chrominance NR too far, you'll suck the life out of the colors in your image.

Not surprisingly, it's another trade off. In my case, I pushed the L/C NR just a tiny bit past the default to 6 and 6 respectively.

You'll be able to see the effect of chrominance NR by looking at the flat colored grey wall in the background. Just don't forget to check other areas of your image with the settings you choose. For me it was a close look at her iris, where pushing the chrominance NR too far lost some of the beautiful colors in her eye.

Compare the same crop from above with and without Luminance/Chrominance noise reduction applied:

With Luminance & Chrominance NR set to 6 (mouseover to compare w/o)

If you've read my previous article on B&W conversion, you'll know that I don't mind a little noise/grain in my images at all, so this level doesn't bother me in the least. I could chase the noise even further if I really wanted to, but always remember that doing so is going to be at the expense of detail/color in your final result. As with most things in life, moderation is key!

Sharpening

If you are going to sharpen your image a bit, this is probably the best time to do so. The problem is that usually sharpening is the last bit of post-processing you should do to your image, due to it's destructive nature. Plus, lately I've grown accustomed to sharpening by using an extra wavelet scale during my skin retouching in GIMP (you'll see below in a bit).

So, I'll avoid sharpening at this stage. If I was going to use it here at all, it would be just very, very light. Also, if you do any sharpening at this stage, try to make sure that it happens after any noise reduction in the pipeline.

Settings so far (everything else zero)...

Exposure Compensation	2.30
Black	150
WB Temperature	7300
WB Tint	0.545
Impulse NR	55
Luminance NR	6
Chrominance NR	6

Lens Correction

This is actually a section that deserves its own post, detailing methods for correcting for lens barrel distortion with Hugin. RawTherapee actually has an "Automatic Distortion Correction" that will effect pincushion distortion in your images.

In my case, I was shooting at the long end of the lens at 50mm, and the distortion is minimal. So I didn't bother with correcting this (it might have been needed at a shorter focal length, and being closer to the subject, though).

In Summary

That about wraps up the RAW "development" I'm going to do on this image. I try to keep things minimal where possible, though I could have gone further and adjusted color tones and LAB adjustments here as well. In fact, with the exception of Wavelet Decompose for skin retouching, and some other masking/painting operations, I could do most of what I want for this portrait entirely in RawTherapee.

I know that this reads really long, but the truth is that once I am accustomed to a workflow, this takes less than 5 minutes from start to finish (faster if I've already fiddled with other images from the same set). All I really modified here was Exposure, White Balance, and Noise Reduction.

Finally, as I hinted at earlier, here is the final version after doing all of these RAW edits, as we get ready to bring the image into GIMP for further processing:

This is the one to download if you want to follow along in GIMP below.
Just click the image to open in a new window, then save it from there.

GIMP Retouching

Well, here we are. Finally. It's the home stretch now, so don't give up just yet!

If you didn't follow along with the RAW processing earlier, you can download the full resolution JPG output from Raw Therapee by clicking here:

Download the full resolution JPG from RawTherapee.

Armed with our final results from RawTherapee, we're now ready to do a little retouching to the image.

The rest of my GIMP tutorials can be found here:
Getting Around in GIMP

The overall workflow and the order in which I approach them is dependent on my mood mostly. Most times, I enjoy doing skin retouching, so I'll often jump right in with Wavelet Decompose and play around. Really, though, I should start shifting Wavelet Decompose to a later part of my workflow, and fix other things like removing objects from the background and fixing flyaway hairs first.

This way, I can directly re-use wavelet scales for a slight wavelet sharpening while I have them.

Looking at this image so far, I can spot a few broad things that I want to correct, and I'm going to address them in this order:

Touchup flyaway hairs
Crop & remove distracting background elements
Skin retouching with Wavelet Decompose
Contour paint highlights
Apply some color curves

Touchup Flyaway Hairs

If you can have the model bring a hairbrush with them to a shoot - DO IT. Seriously. Your eyes and carpal tunnel will thank me later.

Even with a brush or hairstylist/make-up artist the occasional hair will decide to rebel and do its own thing. This will require us to get down to the details and fix those hairs up.

Luckily for me, Mairis hair mostly cooperated with us during the shoot (and where it didn't I kind of liked it). To illustrate this step, though, I'm going to clean up some of the stray hairs on the left side of the image (the right side of her face).

Luckily for me, the background is a consistent color/texture. This means cloning out these hairs shouldn't be too much of a problem, but there are still some things you should keep in mind while doing this.

Here is the area that I'd like to clean up a little bit:

Sometimes you just have to work one strand of hair at a time...

I will usually use a hard-edged brush because a soft-edge will smear details on its edges, and can often be spotted pretty easily by the eye. This works because the background is relatively constant in grain and color.

I'll sample from an area near the hair I want to remove, and set the brush to be "Aligned". I also try to keep the brush size as small as I can and still remove the hair.

The thing to keep in mind is how the hair is actually flowing, and to follow that. I will often follow outlying strands of hair back to where they start from the head, and begin cloning them out from there.

I also try not to get too ambitious (some stray hairs are sometimes fine). Removing too many at once can lead to unrealistic results, so I try to be conservative, and to constantly zoom out and check my work visually.

Try not to leave hairs prematurely cut off in space if possible, it tends to look a bit distracting. If you want to remove a hair that crosses over another strand that you may want to keep, make sure to adjust the source of the clone brush so you can do it without leaving a gap in the leftover strand.

Here is a quick 5 minute touchup of some of the stray hairs (mouseover to compare to the original):

Occasionally, you'll need to fix hairs that are crossing over other hair (sort of like a virtual "brushing" of the hair). In these cases, you really have to pay careful attention to how the hair flows and to use that as a guide when choosing a sample point with either the clone or heal brush.

If this sounds like a lot of work - it is. Thankfully, once you've become accustomed to doing it, and doing it well, you'll find yourself picking up a lot of speed. It's one of those things that's worth learning to do right, and to let practice speed it up for you.

I actually like the cascading hair around her face opening up to a pretty color, so that's about as far as I'm going to go with stray hairs on this image.

Fixing the Background & Cropping

With the limited space I had to shoot this portrait, it's no surprise that I had gotten some undesirable background elements, like the window edges.

There's a couple of ways I could go about fixing these - I could fix the background in place, or I can crop out the elements I don't want.

In my final version shown in the previous post, I wanted to crop tighter, so it worked out well to remove the window on the left. To illustrate how we can remove the window, I'm going to leave the aspect ratio as it is, and walk through removing the distracting background elements.

Removing Background Elements

Because most of the background is already a (relatively) solid color, this isn't too hard. There's just a couple of simple things to keep in mind.

The way I'm going to approach this is to make a duplicate of my current layer, and to move the duplicate into place such that the background will cover up parts of the window I want to remove. Then I'll mask the duplicate layer to hide the window.

I start by choosing an area of the background that's similar in color/tone:

Thankfully the background is relatively consistent.

I'll then move the duplicate layer so that the green area covers up the window to the left:

Position the duplicate layer so the green area now covers up the window

Here is what this looks like in GIMP, with the duplicate layer set to 90% opacity over the base layer (so you can see where the window edge is):

Moving the duplicate layer over to cover the window

Now I'll add a black (fully transparent) layer mask over the duplicate layer, and I'll paint white on the mask to cover up the window edge (with a soft-edged brush). This give me results that look like this:

After applying a transparent mask, and painting white over the window edge

The problem is that the background area from the duplicate is a bit darker than the base layer background, and the seam is visible where they are masked. To fix this, I can just adjust the lightness of the duplicate layer until I get a good match.

I used Hue-Saturation to adjust the lightness (because I wasn't sure if I would need to adjust the hue slightly as well - turns out I didn't). I found that increasing the Lightness value to 3 got me reasonably close:

After increasing duplicate layer Lightness to 3

To further fix the lower part of the window, I just repeated all the steps above with another duplicate of the base layer, just shifted to cover the lower part of the window. I had to mask along her sweater. Here is the result after repeating the above steps:

After repeating above steps for the lower left corner

The results are ok, but could be just a little bit better. Visually, the falloff of light on the background doesn't match what's happening on her body, so I added a small gradient to the lower left corner to give it a more natural looking light falloff:

Adding a gradient to the lower left background helps it look more natural

Fixing the slight window/shadow on the right is easily done with a clone/heal tool combination. The final result of quickly cleaning up the background is this:

Finished cleaning up the background

I could have spent a little more with this, but I'm happy with the results for the purpose of this post. If your cloning efforts leave obvious transitions between tones, the Heal tool can be helpful for alleviating this (especially when used with large brush radii, just be prepared to wait a bit).

With the background squared away, we can move on to one of my favorite things to play with, skin retouching!

Skin Retouching with Wavelet Decompose

I had previously written about using Wavelet Decompose as a means for touching up skin. As I said in that post, and will repeat here:

The best way to utilize this tool is with a light touch.

Re-read that sentence and keep it in mind as we move forward.
Don't make mannequins.

Ok, with a layer that contains all of the changes we've made so far rolled up, we can now decompose the image to wavelet scales. In my case I almost always use the default of 5 scales unless there's a good reason to increase/decrease that number.

For anyone new to this method, the basic idea of Wavelet Decompose is that it will break down your images to multiple layers, each containing a specific set of details based on their relative size, and a residual layer with color/tonal information. For instance, Wavelet scale 1 will contain only the finest details in your image, while each successive scale will contain larger and larger details.

The benefit to us is that these details are isolated on each layer, meaning we can modify details on one layer without affecting other details from other layers (or adjust the colors/tones on the residual layer without modifying the details).

Here is an example of the resulting layers we get when running Wavelet Decompose:

Wavelet scales from 1 (finest) to the Residual

After running Wavelet Decompose, we'll find ourselves with 6 new layers: Residual + 5 Wavelet scales. I am going to start on Wavelet scale 5.

If you hold down Shift and click on a layer visibility icon, you'll isolate just that single layer as visible. Do this now to Wavelet scale 5, and let's have a look at what we're dealing with.

I usually work on skin retouching in sections. Usually I'll consider the forehead, nose, cheeks to smile lines, chin, and upper lip all as separate sections (trying to follow normal facial contours). Something like this:

Rough breakdown of each area I'll work on separately

I'm going to start with the forehead. I'll work with detail scales first, and follow up with touchups on the residual scale if needed to even out color tones. Here is what Wavelet scale 5 looks like isolated:

Forehead, Wavelet scale 5

It may not seem obvious, especially if you don't use wavelet scales much, but there's a lot of large scale tonal imperfections here. Look at the same image, but with the levels normalized:

These are the tones we want to smooth out

Normalizing the wavelet scale lets you see the tones that we want to smooth out.

My normal workflow is to have all of the wavelet scales and residual visible (each of the wavelet scales has a layer blending mode of Grain Merge). This way I'm visually seeing the overall image results. Then I will select each wavelet scale as I work on it.

I'll normally use the Free Select Tool to select the forehead. I'll usually have the Feather edges option turned on, with a large radius (maybe 1% of the smallest image dimensions roughly - so ~35 pixels here). Remember to have your layer selected that you want to work on.

With my area selected, I'll often run a Gaussian Blur (IIR) over the skin to smooth out those imperfections. The radius you use is dependent on how strong you want to smooth the tones out. Too much, and you'll obliterate the details on that scale, so start small.

Here is my selection I'll work with (remember - my active layer is Wavelet scale 5):

Forehead with selection (feather turned on to 35px)

Now I'll experiment with different Gaussian Blur radii to get a feel for how it will effect my entire image. I settled on a high-ish value of 35px radius, which gave me this as a result (mouseover to compare to original):

Forehead, Wavelet scale 5 after Gaussian Blur (IIR) 35px radius

Just with this small change to a single wavelet scale, we can already see a remarkable improvement to the underlying skin tones, and we haven't hurt any of the fine details in the skin!

In some cases, this may be all that is required for a particular area of skin. I could push things just a tiny bit further if I wanted by working globally again on a finer wavelet scale, but I've learned the hard way to back off early if possible.

Instead, I'll look at specific areas of the skin that I may want to touch up. For instance, the two frown lines in the center of the forehead. I may not want to remove them completely, but I may want to downplay how visible they are. Wavelet scales are perfect for this.

Small frown lines I want to reduce

Because each of the Wavelet scales are set to a layer blend mode of Grain Merge, this means that any area that has a completely grey color will not effect the final image. This means that you can paint with medium grey RGB(128,128,128) to completely remove a detail from a layer.

You can also use the Blur/Sharpen brush to selectively blur an area of the image as well. (I've found that the Blur tool works best at smaller wavelet scales - it doesn't appear to make a big difference on larger scales).

So, if we look at Wavelet scale 5 where the frown lines are, we'll see there's not much there - it was already smoothed earlier. If we look at Wavelet scale 4 though, we'll see them prominently.

I'll use the Heal Tool to sample from the same wavelet scale in a different location, and paint over just the frown lines. I'll work on Wavelet scale 4 first. If needed, I can also move down to Wavelet scale 3 and repeat the same procedure there.

A couple of quick passes just over the frown lines, and the results look like this:

Cloning over frown line on scale 4 & 3 (mouseover to compare)

I could continue over any other blemishes I may want to correct, but small individual blemishes can usually be fixed with a little spot healing quickly.

Moving on to the nose, the tones have different requirements. Overall, the tones on Wavelet scale 5 are similar to the forehead. In this case, a similar amount of blurring as the forehead on scale 5 will nicely smooth out the tones. Here is the nose after a slight blurring (mouseover to see original):

Nose with 35px Gaussian blur on Wavelet scale 5
Mouseover to compare to original

There is a bit of color in the nose that is slightly uneven that I'd like to fix. This is relatively easy to do with wavelet scales, because I can modify the underlying color tones of the nose without destroying the details on the other scale layers.

In this case, I'll work on the Wavelet residual layer.

I'll use a Heal Tool with a large, soft brush. I'll sample from about the middle of the nose, and clean up the slightly redder skin by healing new tones into that area. I'll follow the contours of the nose and the way that the light is hitting it in order to match the underlying tones to what is already there.

After a little work these are the results (mouseover to compare to original):

Healing on the Wavelet residual scale to even tones
Mouseover to compare to original

Next I'll take a look at the eyes and cheek on the brighter side of her face.

Overall tones are good here, just some slight retouching required

The tones here are not bad, particularly on scale 5. After making my selection, I've applied a blur at 25px just to smooth things a bit.

A slight 25px blur to smooth overall tones
Mouseover to compare to original

The dark tones under/around the eyes is a bit different to deal with. As before, I'll turn to working on the Wavelet residual layer to brighten up the color tones under the eyes.

I use the Heal Tool to sample from a brighter area of skin near the eye. Then I'll carefully paint into the dark tones to brighten them up, and to even the colors out with the surrounding skin.

Carefully cloning/healing brighter skin tones under the eyes
Mouseover to compare to original

Wavelets are amazing for this type of adjustment, because I can brighten up/change the skin tones under the eyes without effecting the fine skin details here like small wrinkles and pores. The textual character remains unchanged, but the underlying skin tones can be modified easily.

The same can be done for the slightly red tones on the cheek, and at the edge of her jaw. Which I did.

I'm purposefully not going to modify the fine wrinkles under the eyes, either. These small imperfections will often bring great character to a face, and unless they are very distracting or bad, I find it best to leave them be.

A good tip is that even though these small imperfections may seem large when you're pixel peeping, get into the habit of zooming out to a sane zoom level and evaluate the image then. Sometimes you'll find you've gone too far, and things begin to creep into mannequin territory.

Don't make mannequins!

In Summary

This entire post is getting a little long, so I'm going to stop here with the skin retouching breakdown.

Also, that's honestly about it as far as the process goes. Just repeat on the areas that are left (right cheek, chin, and upper lip). You can just apply the processes I described above to those other areas, in the same way.

To summarize, here are the tools/steps I'll use with Wavelet Decompose to retouch skin:

Area selection with Gaussian blur to even out overall tones at a particular scale
Paint with grey, Clone, Heal on wavelet scales to modify specific details
Clone/Heal on wavelet residual scale to modify underlying skin tones/colors (but leave details intact)

Here are the final results after using only Wavelet Decompose (mouseover to compare to original):

After retouching in Wavelet Scales only
Mouseover to compare to original

Spot Touchups

There may be a few things that still need a little spot touchup that I didn't bother to mess with in Wavelet scales.

In my case, I'll clone/heal out some small hairs along the jaw line, and touch up some small spots of skin individually. This is really just a light cleaning, and I usually do this at the pixel level (obnoxiously zoomed in, and small brush sizes).

I also use a method for checking the skin for areas that I may want to touchup, but might not be immediately visible or noticeable. It uses the fact that the Blue channel of an image can show you just how scary skin can look (seriously, color decompose any image of skin, and look at the blue channel).

I want to outline the idea, but this post is going on far too long already. I'll follow up this post immediately with the method of using the Blue channel as a "check layer" (Calvin Hollywood recently illustrated the idea in Ps).

You can find the detailed tutorial for "check layers" here at this post now.

Contour Painting Highlights

One of the downsides of using Wavelet scales for modifying skin is that if you're blurring on some of the scales, you'll sometimes decrease the local contrast in your image. This isn't so bad, but you may want to bring back some of the contrast in areas you've touched up.

I'll utilize something I refer to as "Contour Painting" (I had previously written about using this method for enhancing a pinup image).

What I'm going to do is basically add some transparent layers over my image, and set their layer blend modes to "Overlay".

Then I'll paint white over contours I want to enhance, and adjust the opacity of the layer to taste. (This is highly subjective, so I'm going to just show a quick idea of how I might approach it - you can get as nuts with this as you like...).

Here I've added a new transparent layer on top of my image, and set the Layer Blend Mode to Overlay. Then I painted white onto contours that I want to highlight:

Painting on the Overlay layer along contours to highlight

It looks strange right now, but I'll add a large radius Gaussian Blur to smooth these tones out. I used a blur radius of 111 pixels. Here is what it looks like after the blur:

Blurring the Overlay layer with Gaussian Blur (111 pixel radius)

Finally, I'll adjust the opacity of the Overlay layer to taste. I'll usually dial this way, way down so that it's not so obvious. Here, I've dialed the opacity back to about 20%, which leaves us with this (mouseover to compare):

After setting the Overlay layer to 20% opacity (still a little high for me, but it's good for illustration)
Mouseover to compare to original

I will sometimes add a few more of these layers to enhance other parts of the image as well. I'll use it (reall, really, really lightly!!!) to enhance the eyes a bit, and in this case, I used an even larger layer to add some volume and highlights to her hair as well.

Here is the results after adding some eye and hair highlight layers as well (mouseover to compare no highlights):

Face, eyes, and hair contour painting result
Mouseover to compare to original

Color Curves

Finally, I like to apply some color curves that I have around and use often. I've been heavily favoring a Portra emulation curve from Petteri Sulonen that he calls Portra-esque, especially for skin. It has a very pretty rolloff in the highlights that really renders pretty colors.

If I feel it's too much, I can always apply it on a duplicate of my image so far, and adjust opacity to suit. Here is the same image with only the Potra-esque curve applied:

Image so far, with a Portra-esque color curve applied
Mouseover to compare to original

If you're curious, I had written up a much more in-depth look at color curves for skin here: Getting Around in GIMP - More Color Curves (Skin). You can actually download the curves for Portra, Velvia, Provia emulation on that page.

Sharpening

Finally. The last step before saving out our image!

For sharpening, I actually like to use one of the Wavelet scales that I generated earlier. I'll just duplicate a low scale, like 2 or 3, and drag it on top of my layer stack to sharpen the details from that scale.

In this case, I liked the details from Wavelet scale 2, so I duplicated that layer, and dragged it on top of my layer stack. The blend mode is already set to Grain Merge, so I don't have to do anything else:

Wavelet scale 2 copied to the top of the layer stack for sharpening
Mouseover to compare to original

Finally at the End

If you're still with me - you really deserve a medal. I'm sorry this has run as long as it has, but I wanted to try to be as complete as I could.

So, for a final comparison, here is the image we finished with (mouseover to compare to what we started with before retouching in GIMP):

Our final result
Mouseover to compare to original (from RT)

Not too bad for a little bit of fiddling, I think! I know that this tutorial reads really, really long, but I promise that once you've understood the processes being used, it's actually very quick in practice.

I hope that this has been helpful to you in some way! If you happen to use anything from this tutorial please share it. I'd love to see what others do with these techniques.

If you appreciate any of the stuff I'm writing, please consider donating a little (even 50¢ is great), or possibly clicking something else on the page... :)
If not, would you consider linking/favoriting/+1/whatever you kids do these days and let others know about it? I would seriously appreciate it!

The Open Source Portrait (Equipment & Environment)

2013-03-20T15:34:00.000-05:00

As promised, I thought it might be fun/helpful to walk through a portrait shooting session from start to finish using only Free/Open Source Software (F/OSS). This is my personal workflow, and hopefully I can do the subject some justice.

A Model Idiot.

I'm going to walk through the creation of a portrait of my friend Mairi that I did not too long ago. I had been wanting to shoot with Mairi for a while, and only recently got a chance to sit down and actually do it quickly. For reference, here is one of the final results that I'll be walking through the process for:

Mairi 6 - possibly my favorite from the shoot.

I say "quickly" because this image was almost spur-of-the-moment, shot in a less than ideal location with about 10 minutes of setup to dial everything in (oh - and was shot while I was keeping an eye on my 2 year old daughter who was running around playing).

The next tutorial in this series is here:
The Open Source Portrait (Postprocessing)

The Idea

Before we even sat down, I already had a rough idea of how I wanted to light this portrait. I knew I wanted a big, diffuse key light set up to be close to a Rembrandt style of lighting, and that I wanted to possibly include a second light for fill to brighten up the shadows just a bit.

Samples of Rembrandt paintings.

This is actually a common theme for me, as I had built a softbox that I could hold in my left hand, while shooting with my right. Because I had been using wider angle lenses for a long time, it became a sort of habit for me to light in this manner (just look at the majority of the shots in my Off Camera Flash set on Flickr to see what I mean).

If you remember my previous post, Visualize Photography Lighting Setups in Blender, you'll see a rough initial idea based around a softbox at ~60° elevation, and ~60° off axis to camera left (300° frame from below).

Virtual lighting diagram - see 300°.

The only difference this time was that I knew from experience that a single softbox will leave one side of my subject quite dark, so I wanted to try filling in those shadows with my second flash (they could also have been filled with a reflector as well).

The Equipment

I don't really have much in terms of equipment that is specifically for shooting portraits. This is the gear I use when I'm doing off-camera flash work.

Almost all of my photography gear...

I know it might seem redundant, but for the sake of completeness and to document this as well as I can, I'm going to walk through the list of what I'm using.

The product links below will take you to their respective Amazon page.
Just to be clear, I am using my affiliate code in these links. This means that if you click a link and buy anything on Amazon afterwards, I'll receive a percentage of the sale. This actually doesn't affect anything for you (the price is still the same, etc.) - just that you'll be supporting me if you do.
I just wanted to let you know, so you can make the choice to click (and if you do - thank you!).

Camera

My new toy from this past Christmas is the Olympus OM-D E-M5. I got mine with the Olympus 12-50mm F3.5-6.3 because, seriously - if I'm going to get a weatherproof camera, I might as well have a lens with the same protection (as part of a kit it's a good price - I probably wouldn't have bought this lens separately, I'm more of a prime shooter normally).

The camera is not the important part of the equation here, though. More important is getting a focal length that would produce pleasing compression without flattening things out too much. Here's a good comparison of what different focal lengths will do to the perspective of your subject from photographer Stephen Eastwood:

Comparison of compression effects from focal lengths. (via Stephen Eastwood)

With my old 17mm pancake lens, I would be looking at distortion like the 2^nd row middle, but at the long end of my zoom (50mm), I'm shooting more for 1^st row, far right. Traditionally, you'll want to be around 85mm (short-telephoto) or higher for portraits.

Speedlights/Flashes

I wanted to learn to do off-camera lighting from the basics, and to work my way up from there. If you're interested in doing something similar, I can't recommend David Hobby's blog Strobist enough.

He even has an amazing series of tutorials to walk you through the basics of learning to use flashes called Lighting 101/2. The entire series is well worth the time to read through.
Seriously. Go read it.

With that being said, I knew I only needed a dependable manual flash to learn with (no need for TTL metering/control for me). In checking around, it appeared that the best price/performance ratio I could find for a manual flash was the Yongnuo YN-560 Speedlight Flash. At the time I bought mine, they were only about $75 USD, and are a great value (I've never had a problem with reliability on mine).

(This flash may have been replaced with the Yongnuo YN-560 II, which is identical with the exception of the LCD screen on the back in lieu of LED status lights).

One of the nice features of this flash is that it does have an optical slave, meaning that it will fire when it detects another flash firing nearby. This is how I got my fill flash to fire without needing another set of radio triggers.

Speaking of radio triggers...

Wireless Triggers

Of course, the speedlights and camera wouldn't work well together without some way for the camera to trigger the flashes. A few years back my friend Sean gave me his old Cactus v4 triggers as a hand-me-down, so I decided to stay with the brand when I picked up some new ones. I went with these, the Cactus V5 Wireless Flash Triggers.

You don't have to have a wireless trigger to fire the flash, you could also go the wired route using a PC-sync cord (the YN560 will accept these, you'll just need a hotshoe-sync cord adapter if your camera doesn't already have a PC jack), or if your camera/flash system are integrated (built-in commander mode, etc.).

You could conceivably also use these with the pop-up flash on the camera. Just set the camera flash to the lowest possible setting, and use it to activate the Yongnuo in slave mode.

Lighting Modifier (DIY Softbox)

I'm firmly in the Do-It-Yourself camp when it comes to lighting modifiers (or I'm cheap - read it how you want). The biggest problem is that costs can spiral out of control if you get a nasty case of GAS (Gear Acquisition Syndrome), and I'd rather spend money on things I can't easily build myself. (This list of softboxes on Amazon shows most 24" softboxes in the +$30 USD range).

Luckily, a simple softbox is something I can build myself...

Here's another view of my DIY softbox (with measurements):

This was constructed with foam-core board that can be picked up just about anywhere for pretty cheap. I believe I paid something like $7 USD for each board, and needed 2 to cut all my pieces.

I only had to cut 4 identical trapezoids (24" base x 16" height x 4" top), and then a single 4"x4" square for the end. The whole thing is held together with clear packing tape. I traced my flash head on the square piece and cut out with a simple X-Acto knife.

The fabric on the front was just an old white sheet I had laying around, but any lightweight, white fabric can do. This was also just taped to the front. You could get fancy and tape a tissue inside as a second diffusion panel, but I didn't bother.

Udi over at DIY Photography has a great post detailing how to build 24 different softboxes if you want other inspirations.

The Environment

I envy photographers sometimes when I get a chance to see these wonderful studio spaces they have access to for shoots. Maybe if I was doing this as part of my income, then I could justify the expense required for those spaces, but instead this is purely a hobby. Oh, and check the previous section about me being cheap.

With that in mind, this is the space I had available to shoot my portrait:

The only space to shoot in. I like a challenge!

This is actually a very pretty corner to shoot in during the late afternoon. The sun is just behind that corner, and gets diffused nicely in this spot. If the subject is back in the corner, there's a nice ratio of light coming from the windows on either side to produce pleasing light.

You'll notice a microphone stand in the image. That became my impromptu light stand when I taped the softbox to the side of it. I had Mairi sit in the chair, and then adjusted the height/angle of the box before taping it in place.

The small blue bucket on the desk is one of the places I had originally set my bare flash down for fill light.

The First Pass

Initially, I had Mairi backed into the corner, and the lights arranged around her. This didn't work out as I had hoped.

The reason was that the spill light from the softbox and bare flash were casting shadows on the walls that I didn't like. The relative exposure of the corner was also pretty bright compared to my subject, which didn't really give much separation at all from the background.

So how to fix this? Well, Strobist has a nice lesson about flash positions and their relative effect on the resulting image (scroll down to the "Light Position: Distance" section). In particular, the ratio between the flash-to-subject distance, and the flash-to-background distance. When they are close, you'll get a background exposed similar to your subject.

Here, have a look at one of the images from that lesson, that shows the light being about the same distance from the subject as the subject from back wall:

Flash to subject ~5 feet, with the wall ~6 feet behind the subject.

Strobist shows this great example below of changing this ratio for a different effect:

~1:1 ratio on left, ~7:1 ratio on right

On the left, David moved the flash to be ~25 feet from the subject, with the wall about 31 feet from the flash (approximately a 1:1 distance ratio). After adjusting aperture to get the correct exposure for the subject, you'll see how light the rear wall is.

On the right, the flash was pushed in very close to the subject, ~1 foot. While the rear wall was about 7 feet away from the flash. Once the aperture is adjusted to correctly expose the subject you can see how dark the rear wall became. The ratio is now about 7:1.

This is basically the Inverse Square Law for light intensity (don't click the link unless you really, really want to read more about it).

So with that in mind, I corrected my problems by simply moving everything away from the corner. Honestly, this was just trial and error until I started seeing the rear wall being exposed about where I wanted it relative to Mairi.

The Second Pass

With Mairi moved out of the corner, and the flashes still maintaining the same relative distance to her, I was able to drop the exposure of the corner walls down a bit, and I avoided any strange shadows as well. How simple! (You can see my first pass tries in the contact sheet towards the end of this post).

Here is approximately what my final setup looked like:

All positions are approximate from my memory, tweak as needed. :)

The keylight is the softbox, which is about 18 inches from the models face (just barely out of frame of the camera).

The keylight is set to about 1/2 power (these flashes are GN58 by the way). Remember, if you can get the diffuse light source closer, it will have a much larger apparent size (if it's far away, it will begin to act like a bare flash again - harsh shadows).

The bare fill flash on the table to the camera right is set to about 1/4 power, and has a zoom head that is set to about 100mm. If you're following along, you'll just need to adjust these power ratios for a proper exposure, but those are my settings if it helps to get you started.

I know this probably reads a bit long, but the truth is that all the steps to get to a final lighting position setup took about 5-10 minutes total. Keep in mind, this was my first time lighting like this in the space, and any future shoots will be even faster (and cleaner once I put away all the kids toys and clutter!).

The Shoot

Once I had the lighting and positioning dialed in, the shoot was a (relative) breeze. I had a couple of things going for me that made it so easy, though. Mairi is a friend of mine, and this was much more like hanging out and joking while I occasionally snapped a photo. She has also modeled in the past, so was absolutely comfortable in front of the camera, and knew how to move and pose for different shots already.

What if the model is not already a friend and/or experienced, though?

This section is not intended to be comprehensive, more of my ramblings around my general thoughts at the time. I'm not covering some essentials like model releases and other things.
To be honest, this section (poses/directions/etc.) deserves a writeup of it's own.

Mood/Music/Atmosphere

I think that the overall atmosphere and mood is vitally important to helping the model do their best (and by extension to get good photos). You want them to feel relaxed and at ease (usually).

If you have some way to play music, ask them to bring something they like along. It might also help to let them know what type of shoot you'd like to do, and to plan music appropriately (Vibrant and full of energy? Somber and serious? Easy and lighthearted? Pick music that matches the mood you want to capture).

It's probably also a good idea to cover basic essentials as well, including restroom/changing room for privacy (if needed), as well as refreshments (avoid snacks or have a mirror handy - no need to spend all your time editing food out of teeth - or discoloration).

I'd also spend some time just putting the equipment down and chatting with your subject (if possible). Build a rapport and get to know each other a bit before you even pick up the camera.

Plan

It will help you immensely if you have at least a rough idea of what types of shots you'd like to try to get ahead of time. Communicate this to your subject if possible before the shoot as well. Having examples of poses and other visual aids will be helpful as well (other photos, magazine clippings, etc. for instance).

For Mairi, I only actually had two shots in mind up front:

The only two poses I had anything to do with.

Every other shot was her moving through various poses without any direction from me except for the occasional request to direct her gaze somewhere.

Shoot

Don't forget to keep talking with the subject (put yourself in their shoes - just sitting there while you shuffle around clicking away and no feedback kind of sucks). Direct them as needed. And don't forget to come out from behind the lens occasionally.

If you want them to do something specifically, try to show them if possible. Don't touch them. (Seriously, don't). If you absolutely have to adjust a hair or piece of clothing, make sure to ask them first.

Finally, a couple of tidbits I picked up from around the web that I thought were neat.

If you want to try and get a natural look, try having the subject look down for a moment and then ask them to bring their head and eyes up to position (asking them to look directly into the lens can capture some wonderful eyes sometimes, especially if you fire off a shot before their head is all the way up into position).

If you want to relax them and possibly catch some nice candid moments, try asking them to do something silly (like faking a cough).

A simple joke can work wonders here too (keep it clean, don't be a creep).

I know that the final images from this session I produced were all more somber, but of all the shots I took, most of them had her laughing and cracking up from some stupid comment/joke I made during the shoot.

So you can get a better feel of how the shoot went, here's a couple of sample contact sheets (chronologically ordered):

Mairi in a dark sweater (notice the first 6 shots are close to the corner).

Mairi in a light grey sweater.
By the time we got to these, I had the lighting dialed in for what I wanted.

You can clearly make out in the first contact sheet my first pass lighting with Mairi much closer to the corner. I didn't like the spill lights on the background, so moved her out of the corner more, which evened out the background, and darkened it a bit for me.

In Summary

Whew! That ran longer than I thought it would. If you've read this far, I applaud you.

So, that's basically it for the actual shoot. We worked through a few different poses, I snapped a bunch of photographs, and we had a few laughs. The next part will deal with what comes next - culling all those shots to the keepers, and deciding how to approach post-processing those into portraits I like.

Also, I think I may just go ahead and make a RAW file available in the next session so anyone following along at home has something to work with.

In the next post, I'll be developing my image from my RAW file in Rawtherapee, bringing that image into GIMP for retouching, and finally exporting the end result.

If you want to get a head start on some of the things I'll be talking about in the next post, check out:
Getting Around in GIMP - Skin Retouching (Wavelet Decompose)

I hope this has been at least interesting for you, and remember to check back in a bit for the last part of this Open Source Portrait walkthrough!

The next tutorial in this series is here:
The Open Source Portrait (Postprocessing)

Calvin Hollywood Freaky Details in GIMP

2013-02-26T17:07:00.000-06:00

German photographer/digital artist/photoshop trainer Calvin Hollywood has a rather unique style to his photography. It's a sort of edgy, gritty, hyper-realistic result, almost a blend between illustration and photography.

As part of one of his courses, he talks about a technique for accentuating details in an image that he calls "Freaky Details".

The rest of my GIMP tutorials can be found here:
Getting Around in GIMP

Here is the original video I saw him describing this technique in from Scott Kelby's blog, Photoshop Insider during a guest appearance:

[Update]
I've been told there are problems loading the page with IE if I include the video I first saw this technique on. So, to make sure everyone can see this tutorial, the original video is on this page towards the end (it's the only video on the page)...

Go check it out, then come back to read the rest of the tutorial!

And here is a more current Youtube video from Calvin that describes this technique using a different image:

In my meandering around different retouching tutorials I came across it a while ago, and wanted to replicate the results in GIMP if possible. There were a couple of problems that I ran into for replicating the exact same workflow:

Lack of a "Vivid Light" layer blend mode in GIMP
Lack of a "Surface Blur" in GIMP

Those problems have been rectified (and I have more patience these days to figure out what exactly was going on), so let's see what it takes to replicate this effect in GIMP!

Replicating Freaky Details

Requirements

The only extra thing you'll need to be able to replicate this effect is G'MIC for GIMP.

You don't technically need G'MIC to make this work, but the process of manually creating a Vivid Light layer is tedious and error-prone in GIMP right now. Also, you won't have access to G'MIC's Bilateral Blur for smoothing. And, seriously, it's G'MIC - you should have it anyway for all the other cool stuff it does!

Summary of Steps

Here's the summary of steps we are about to walk through to create this effect in GIMP:

Duplicate the background layer.
Invert the colors of the top layer.
Apply "Surface Blur" to top layer.
Set top layer blend mode to "Vivid Light".
New layer from visible.
Set layer blend mode of new layer to "Overlay", hide intermediate layer.

There are just a couple of small things to point out though, so keep reading to be aware of them!

Detailed Steps

I'm going to walk through each step to make sure it's clear, but first we need an image to work with!

As usual, I'm off to Flickr Creative Commons to search for a CC licensed image to illustrate this with. I found a great photo to use in this tutorial from Mark Shaiken (Shakeskc) on Flickr - a wonderful portrait of a firefighter!

After a Call by Mark Shaiken on Flickr. (cc-by-nc-sa)

This is a great image to test the effect, and to hopefully bring out the details and gritty-ness of the portrait.

1./2. Duplicate background layer, and invert colors

So, duplicate your base image layer (Background in my example).

Layer → Duplicate
(Shift-Ctrl-D,Shift-⌘-D)

I will usually name the duplicate layer something descriptive, like "Temp" ;).

Next I'll just invert the colors on this "Temp" layer.

Colors → Invert

So right now, we should be looking at this on our canvas:

The inverted duplicate of our base layer.

What the Layers dialog should look like

Now that we've got our inverted "Temp" layer, we just need to apply a little blur.

3. Apply "Surface Blur" to Temp Layer

There's a couple of different ways you could approach this. Calvin Hollywood's tutorial explicitly calls for a Photoshop Surface Blur. I think part of the reason to use a Surface Blur vs. Gaussian Blur is to cut down on any halos that will occur along edges of high contrast.

There are three main methods of blurring this layer that you could use:

Straight Gaussian Blur (easiest/fastest, but may halo - worst results)

Filters → Blur → Gaussian Blur
Selective Gaussian Blur (closer to true "Surface Blur")

Filters → Blur → Selective Gaussian Blur
G'MIC's Smooth [bilateral] (closest to true "Surface Blur")

Filters → G'MIC → Enhancement → Smooth [bilateral]

I'll leave it as an exercise for the reader to try some different methods and choose one they like. (At this point I personally pretty much just always use G'MIC's Smooth [bilateral] - this produces the best results by far).

For the Gaussian Blurs, I've had good luck with radius values around 20% - 30% of an image dimension. As the blur radius increases, you'll be acting more on larger local contrasts (as opposed to smaller details) and run the risk of halos. So just keep an eye on that.

So, let's try applying some G'MIC Bilateral Smoothing to the "Temp" layer and see how it looks!

Run the command:

Filters → G'MIC → Enhancement → Smooth [bilateral]

The values I used in this example for Spatial/Value Variance.

The values you want to fiddle with are the Spatial Variance and Value Variance (25 and 20 respectively in my example). You can see the values I tried for this walkthrough, but I encourage you to experiment a bit on your own as well!

Now we should see our canvas look like this:

Our "Temp" layer after applying G'MIC Smoothing [bilateral]

Layers should still look like this.

Now we just need to blend the "Temp" layer with the base background layer using a "Vivid Light" blending mode...

4./5. Set Temp Layer Blend Mode to Vivid Light & New Layer

Now we need to blend the "Temp" layer with the Background layer using a "Vivid Light" blending mode. Lucky for me, I'm friendly with the G'MIC devs, so I asked nicely, and David Tschumperlé added this blend mode for me.

So, again we start up G'MIC:

Filters → G'MIC → Layers → Blend [standard] - Mode: Vivid Light

G'MIC Vivid Light blending mode, pay attention to Input/Output!

Pay careful attention to the Input/Output portion of the dialog. You'll want to set the Input Layers to All visibles so it picks up the Temp and Background layers. You'll also probably want to set the Output to New layer(s).

When it's done, you're going to be staring at a very strange looking layer, for sure:

Well, sure it looks weird out of context...

The layers should now look like this.

Now all that's left is to hide the "Temp" layer, and set the new Vivid Light result layer to Overlay layer blending mode...

6. Set Vivid Light Result to Overlay, Hide Temp Layer

We're just about done. Go ahead and hide the "Temp" layer from view (we won't need it anymore - you could delete it as well if you wanted to).

Finally, set the G'MIC Vivid Light layer output to Overlay layer blend mode:

Set the resulting G'MIC output
layer to Overlay blend mode.

The results we should be seeing will have enhanced details and contrasts, and should look like this (mouseover to compare the original image):

Our final results (whew!) (mouseover to compare original)

Summary

Well, that feels a bit longer than I had anticipated (maybe these just feel longer when I'm writing them?).

This is not an effect for everyone. I can't stress that enough. It's also not an effect for every image. But if you find an image it works well on, I think it can really do some interesting things. It can definitely bring out a very dramatic, gritty effect (it works well with nice hard rim lighting and textures).

For instance, because I couldn't help myself, I color graded the results just a bit to something I like a bit more (muted colors, portra-esque curves):

I thought a slightly desaturated version suited the subject... (mouseover to compare original)

Finally, a BIG THANK YOU to David Tschumperlé for taking the time to add a Vivid Light blend mode in G'MIC.

Try the method out and let me know what you think or how it works out for you! And as always, if you found this useful in any way, please share it, pin it, like it, or whatever you kids do these days...

And if you want more from Calvin Hollywood, he has a book available on Amazon that covers other techniques as well (in Photoshop, of course):

Calvinize: Signature Techniques of Photoshop Artist Calvin Hollywood

A Quick Look at Quality (Google+ vs. Flickr)

2013-02-25T15:35:00.001-06:00

I had recently taken some photos of my friend Mairi, and wanted to share them. Lately I've been using Google+ more and more (they just make it too easy to use), and photographers seem to be one of the larger groups of users.

So I decided to see what my images would look like on Google+ as compared to Flickr (my mainstay), or even 500px. I've been a Flickr user for years (I think the $25USD/year for a Pro account is a pretty good deal for what it offers), and it's been my reference for a while.

So, let's take a look!

Reference

I'm using a recent photo of my friend Mairi as a reference. Specifically, the final 100% jpg output at full size when I was done playing with it in GIMP.

My reference image.

This is the same image I initially uploaded to Google+ and Flickr in full resolution, at 100% JPG quality, optimized, progressive, DCT Floating Point, Subsampling 4:4:4.

I resized the image in GIMP before uploading here to 550px wide, and used Sinc (Lanczos3) resizing (don't worry, I'm just putting this up here for reference), and exported as PNG using RIOT.

Method

So what I did was, after uploading the full quality image in it's final form to both Google+ and Flickr, I then went and re-downloaded the image from the same services. Google+ will resize your images to a maximum side dimension of 2048px for you (unless you upload directly through picasaweb/Picasa).

So I used the 2048px size as the output from both Flickr and Google+ (they both provide a resized version of your image at those dimensions). I brought them into GIMP, and compared them by setting one layer over the other in Difference blending mode.

To emphasize what I was seeing, I ran a simple Colors → Auto → Normalize against the difference results.

Here's the resulting Difference:

Of course, anytime you run an image through a compression algorithm, there are bound to be differences in the results. It's important to keep in mind, too, that the Flickr file at 2048 pixels clocks in at about 1.4 MB in size, while the Google+ version is half that at 745 KB.

Here is a 200% crop of an area of interest from Flickr (mouseover for comparison to Google+):

200% crop from Flickr (mouseover to compare Google+)

Here's another 200% crop of the background grain from Flickr (mouseover to compare Google+):

200% crop of background grain from Flickr (mouseover for Google+ comparison)

Results

To my eye, the extra compression that Google+ is adding to the images is not good. There's some aliasing effects along sharp edges that I don't like, and areas of smooth color + grain fare very poorly. There's also some softening/smearing that occurs at the Google+ compression stage that I don't like.

Take this with a grain of salt, of course. These images are being viewed again at 200%, and pixel-peeping. For instance, here are both images resized to 550 pixels wide (Flickr first, mouseover to compare to Google+):

Flickr image at 550 pixels wide (mouseover to compare Google+)

And here is a simple 100% crop of a similar area of interest from earlier (again, Flickr first, mouseover for Google+ comparison):

I think given the choice I would opt for the Flickr version every time. It's a bigger file, but even at 2048 pixels, it's still not likely to exceed 2 MB for an image. A small price to pay to keep the viewing quality of the image a bit higher...

Mairi

2013-02-22T12:04:00.000-06:00

I finally got a chance recently to quickly photograph my friend Mairi.

Mairi 6 by Pat David on Flickr

She's a beautiful woman who happens to be a keen-eyed photographer as well. The best part, though, is that she has modeled in the past, which made her the perfect person to shoot with!

Seriously. The value of having someone who knows what to do in front of the camera is amazing.

This was the one pose I wanted.
Mairi 4 by Pat David on Flickr

For example, the image above, Mairi 4, was one of two images that I asked her to pose specifically for (Mairi 9 is the other one). All of the others were taken with no direction from me, and just letting her do her thing.

The only other pose I had anything to do with...
Mairi 9 by Pat David on Flickr

This shoot was also fun because it was no stress, just to try some things out. I don't have a studio, but I do have a room with big windows on three sides, and a nice gray colored wall in a corner. When we started, it was still sunny out, and the windows acted as giant, diffuse light sources. Unfortunately, I only managed to get one image I liked shooting in natural light...

Mairi by Pat David on Flickr

I had a rough idea of how I wanted to light the image with flashes, though. I have a pair of Yongnuo YN-560 Speedlight Flashes that I've been using for a while to learn (strobist) lighting with. They are fully manual flashes (which is perfect for me - I like the control for these types of shoots).

Shooting bare flashes is really, really harsh, though. A while back when I got my first flash, I threw together a quick DIY softbox from some foamboard and an old white pillowcase. Pretty much the same build as this one on diyphotography.net. Total cost? About $8. This thing has been worth it's weight in gold for me!

Mairi 2 by Pat David on Flickr

My first test shots were a little too close to background (didn't want the wall to get so much spill light), and there was a bare flash camera right that wasn't dialed in yet.

Still, I liked the result in this image. I thought the juxtaposition of the dark left, brighter right on the rear wall vs. the lighting on her face was visually interesting. So a happy little stroke of serendipity.

Of course, after having recently written a rather long-winded tutorial on B&W conversion, I couldn't help myself while I was working with a moody single light source...

Mairi 3 B&W by Pat David on Flickr

Eventually I got my lighting dialed in to where I wanted it. I moved the Mairi and the softbox (taped to an old microphone stand!) further away from the rear wall in order to darken it. This finally got the results I was looking for...

Mairi 5 by Pat David on Flickr

At this point I just let Mairi do her thing, and she moved easily through a handful of poses for me. The whole thing was very organic and fun! The amazing thing is that we shot all of these images (almost 200!) in about one hour.

Mairi 8 by Pat David on Flickr

I realized a couple of neat things while working on these images. My workflow is pretty much entirely all Free and Open Source Software (F/OSS).

I manage all of my files through regular file management tools and the command line (though there are great pieces of software out there for it). The three main pieces of software I've been using lately are:

Exiftool by Phil Harvey
This is usually because I lose EXIF data between Rawtherapee and GIMP, but also for bulk file manipulations like sorting into folders/etc...
Rawtherapee for developing my RAW files.
(no Darktable build for windows yet, and I like my tools to be consistent across environments)
GIMP - do I really need to say anything here?

I figured that with a fully F/OSS workflow, it might be nice to do another tutorial series on the entire process from setup, to lighting, to shooting, to developing, and finally to retouching in GIMP!

I guess what I mean to say is, the next set of Getting Around in GIMP tutorials will be based on fashion/portraiture workflows and retouching from beginning to end. Stay tuned!

Getting Around in GIMP - Black and White Conversion (Part 5)

2013-02-05T14:36:00.000-06:00

So here we are at the end (not really, but too much more and we'll be just beating a dead horse).

What I feel many of posts become...

My intention at this point is to document some sort of workflow that might prove useful to others (and myself - you'd be surprised at how much I learn writing these posts...). So I'll touch on the final topic I wanted to talk about, blending different bits from various layers and working through a couple of examples.

The rest of my GIMP tutorials can be found here:
Getting Around in GIMP
The rest of the tutorials in this series are here:
B&W Conversion - Part 1 (Desaturate)
B&W Conversion - Part 2 (Channel Mixer)
B&W Conversion - Part 3 (Decompose)
B&W Conversion - Part 4 (Pseudogrey, c2g, Blending)

Bits and Pieces

If you'll recall in Part 3 of this series, I had created a small Script-Fu to automatically decompose an image into various color component decompositions. I think this is a handy script to use if you're still finding your own workflow for grayscale conversions.

What's nice about it is that it will automatically generate a handful of automatic conversions for you as layers on your image. This way, you can visually inspect the results and find areas from each conversion that you like. Then you can use Layer Masks (you can go review my old Layer Masks tutorial to brush up if you'd like) to isolate specific areas and to blend it with others.

Enough talk, let's dive into an example...

Pretty Women

Well, pretty woman, but I couldn't resist a Sweeney Todd reference.

What a sweet looking portrait (IMG_8405a) by Alaskan Dude on Flickr

This image is from Flickr user Frank Kovalchek (Alaskan Dude). In trying to get interesting images for these tutorials, I've searched the Creative Commons licensed images on Flickr, and he shows up a lot. Frank, if you read this - thank you for sharing!

So, I thought this might be a nice image to convert. To get a feel for what is possible quickly, I'm going to run my Color Decompose script, which gives me this:

One of the reasons I chose this image is that it allows us to investigate a couple of points mentioned earlier in this series. I want to keep as high a tonal density as possible in the image to help impart a sense of texture and detail. I also want to highlight some parts of the image for their best qualities.

The Scarf

So in this image, I want to keep the texture and detail of the scarf while at the same time to emphasize and accentuate the quality of the models skin. So I want to inspect these results with an eye towards these qualities (this is all highly subjective, of course, but gives us a good place to start from).

So I ask myself:

"Which of these results produces the best quality/texture in the fabric of the scarf?"

I narrowed it down to the Luma Y709F, Luma Y470F, and HSL-L. Of those, I personally like the look of the Luma Y709F. So this becomes my base layer that I will start building on. (Part of the reason is that aesthetically I want focus on the models face, but still retain texture and detail of the scarf around her).

The Luma - Y709F as a "base" layer - chosen for the fabric texture.

At this point I'm more worried about the tones being there rather than a final result. Remember, we're building up the image here, and can make further corrections or enhancements later.

The Skin

Now I want to consider the model herself. There is fine detail in her skin now, but I'm looking to emphasize her overall. I'd like to get the skin a little brighter and in a higher key, to offset the dark background and the scarf.

I'd also like to get it a little smoother/softer looking. Keeping that in mind, have a look at the different channel decompositions above. Not surprisingly, the Red channel looks quite pretty. This is fairly common that the red channel will be complimentary to (Caucasian) skin. (There's even a trick to using a red channel as an overlay on a color image to "enhance" the skin).

So I'm going to try this and see how I like it. I'll place the Red channel over the Luma Y709F channel, and change the blending mode to Overlay.

Luma Y709F base, with a Red channel over (set to blending mode: Overlay).
(Mouseover for comparison).

My Layers should now look like this:

Visually this appears to have a little more impact, but the skin might be blown out a little too much. I could adjust the opacity on the Red layer at this point to attenuate it's influence until I find a result I am happy with.

Remember, part of the reason this may seem to have more impact could be due to a higher contrast at the moment. Sometimes it's best to stand up and look away from the image for a while before committing to a change...

The problem is that the ratio of adjustment between the scarf and the skin might not be what I want. Adjusting the opacity might reduce the effect on the skin, but at the same time will reduce the effect on the scarf by an equal amount. What I need is some way to apply the effect stronger to the scarf or skin.

Well, this is exactly what Layer Masks are for!

Masks

I could add a mask to the red channel, then paint in a mask by hand that isolated the face, and gave a little less opacity to the scarf. It's a lot of tedious, detailed work.

If only there were already a grayscale mask that isolated the face and scarf a bit. Wouldn't that be great?

Well, scroll back up and look at the RGB - Blue and RGB - Green layers. One of these looks like they would be great at isolating the face/hair from the scarf!

So that's exactly what I'm going to try. I'll copy one of those channels (I'll try the RGB - Green and see how it looks), and apply it as a layer mask to the RGB - Red channel Overlay.

So, select the RGB - Green channel, and copy it.

Edit → Copy

Add a layer mask to the RGB - Red channel. (You can initialize the mask to white.)

Right-Click Red Layer → Add Layer Mask...

The mask should be automatically selected when you create it (white border), if not, click on the mask to select it.

Then paste the copied RGB - Green layer.

Edit → Paste

It should now be a Floating Selection.

Anchor the layer by right clicking on the Floating Selection layer, and choosing Anchor Layer.

When you're done, your Layers should look like this:

Now remember, a layer mask will be more transparent the darker the color in it. The lighter areas will show more of the layer it is applied to. So in this case, lighter areas will allow more of the RGB - Red layer to show, while darker areas will show through to the layer below, Luma Y709F.

Our image at this point should look something like this:

Red Overlay masked with the Green channel (mouseover to compare w/o mask).

What this has done is to isolate the models face from the surrounding scarf a bit. You can now modify the opacity of the layer, or adjust the values of the mask using Levels or Curves to adjust the intensity of the result.

You should notice that if you modify the opacity of the RGB - Red layer, the effect will mainly be on the skin of the model thanks to our mask.

Having a look at the results, you'll notice that the scarf has become a little more flat, and the models face has brightened up. I actually like the depth that the Red overlay layer had on the scarf, and think it's a little strong on the models face.

To test something out, I'm going to try Inverting the layer mask. With the RGB - Red layer mask selected (remember - it will have a white outline in the layers window), I'll just invert the colors with:

Colors → Invert

The result should look like this:

Inverted mask results (mouseover to compare the non-inverted mask).

Here's what the layer palette looks like now:

I actually like this result at this point. The overlay and mask added some nice depth to the scarf fabric, and there is still a bit of nice effect on the skin of the model as well. I could go further and adjust the layer mask levels to decrease/increase the masking on the models skin, but I think we can leave this one as is for the moment.

Let's compare the results to a straight desaturation:

Straight Desaturation (mouseover to compare our final result).

A little fussier than a straight desaturation to get here, but I think the results are much nicer and is visually more interesting. I would probably fiddle a bit with the levels of the layer mask to adjust the contribution from the mask to the face/scarf, but this is a good starting point.

Methuselah

Well, this isn't the actual Methuselah, but it is a similar species of Bristlecone Pine (it might be the exact same species, but I'm no botanist). Once again, Flickr user Frank Kovalchek shows up in a CC-BY search for trees!

Bristlecone pine hanging on for dear life at 10,000 feet by Alaskan Dude on Flickr

As before, I started with decomposing the image to different channels to see if there would be any interesting results I liked. I liked the initial result from Luma - Y709F again, but thought I would take a different path this time.

I want the main focus of this image to really be the texture of the gnarled old tree itself, and secondary the lighting of the day across the ground. Inspecting the results of my decomposition, I like the results of the green channel best on the texture of the tree. It is almost the exact same result as from the Luma Y709F.

Green channel of RGB decomposition

The things I personally don't like about the green channel is that the sky is too bright for me, and the ground could be a bit darker compared to the tree. There's also a nice feeling of light on the ground where the sunlight is reflected to the back right of the tree in the image.

Having found a nice layer for the tree texture, I am now looking for something that represents the ground and sky a little closer to what I'd like. I like a slightly darker sky, and for that the Red channel seems like a good compromise (the Blue channel was a little too noisy for me).

On inspection, I also like the Red channel for it's results on the ground near the tree (slightly darker, and retaining the reflection from the sun).

Red channel of RGB decomposition (mouseover to compare to Green channel)

So what I'd like to do is use the Red channel as a base for the sky and ground, and use the Green channel for the tree itself. Once again, layer masks seem like a good approach!

So I'll set my Green channel above my Red channel on my layer palette, and I'll add a layer mask to the Green channel, initialized to Black (full transparency). This lets all of the underlying Red channel show through.

Red channel, with Green channel over + mask

Now with the Layer Mask active (see the white outline in the above image - if it's not active, click on it), I can paint with a white color to allow that portion of the Green channel to show over the Red channel beneath it. Because I want to use the Green channel just for the tree, that is where I will paint.

There's a couple of ways you could produce the mask. As we saw earlier, I could use another channel and fiddle with the levels a bit to really get a good mask, but in this case I felt like painting. I wanted to talk about this because it's a nice way to visually inspect what your masking will do to the image, and you can tweak it as you go. You could use a different color than white to reduce the effect, as well as adjust brush dynamics for a more "painterly" feel and approach.

Here is my quick painted mask, to illustrate:

It's just a quick mask, don't judge me...

And here is what my layers currently look like:

To compare what we've done so far, here are my results from the steps above:

Final blend of Red/Green channels with mask
(mouseover to compare to straight desaturation)

I could call this one done, though there is a bit of noise in the upper-left corner of the sky from the Red channel. I could fix this by adding another layer masked just for the sky (which would allow me to adjust the levels of the sky relative to everything else to taste). I'll leave that as an exercise for the reader, but a good hint would be to start with the Blue channel to build the mask...

Bonus!

As a bonus, I was smart enough this time to actually save my GIMP workspace while I was fiddling with this tree, so I am offering the .xcf file for anyone to use/follow along:

Download the file from my Google Drive:
Getting Around in GIMP BW 5.bz2
The filetype is .bz2, and should open in GIMP directly with no problems.

Grain

Following the ideas from the phenomenal tutorial by Petteri Sulonen on Digital Black and White, he speaks a bit about grain in B&W images. There are a few different synthetic methods of adding grain to an image, but visually I have also never really been impressed with the results too much.

Petteri was kind enough to make available a grain-field that he processed himself from scanned film. You can apply this as an overlay layer on your image, and adjust opacity to suit. I personally agree with him about grain lending a "structure" to a final image, but it's up to you if you want to include it.

Here is a 100% crop of the image above with the grain field laid over it in Overlay blend mode at 100% opacity:

Petteri's grain field applied (mouseover to compare without)

You can download the full-size png of Petteri's grain field from here:

Download the original size file here

In Summary

I could keep doing this all day long, but I'll stop here. I think there's enough information here to begin experimenting and finding a workflow that works well for you. This is an important point - I'm just showing you the methods I've found for conversion, it's up to you to take that info and to make it your own!

The End

Holy crap. That was a seriously long inspection of B&W grayscale conversion in GIMP! We've come a long way since November 14, 2012, when I started writing these. In my head, I honestly believed I could get the entire topic covered in a couple of weeks (as opposed to almost 3 months!).

If you've followed me all the way through these, I feel like I should send you a merit-badge for patience and perseverance! This final installment puts me at just shy of 10,000 words on B&W conversion alone.

I hope in the end that anyone reading this has gotten some useful information somewhere.

As a final note, if you have found any of these tutorials useful, perhaps consider hitting the "Donate" button above on the right (or here: ) to drop a quarter or two through PayPal to me? I use it to pay for more coffee, to keep me writing and motivated. If you don't want to donate, perhaps visit an ad (sponsor) if it's interesting to you (it's like donating, but not actually paying!).

At the least, if you think these have been helpful tutorials, spread the word! Pin it, +1 it, tweet it, blog it, or whatever else it is that you kids do these days! (There's even some buttons just below to help!)

← Part 4 (c2g, Pseudogrey, Layers)

Focus Stacking Macro Photos (Enfuse)

2013-01-22T15:08:00.001-06:00

I recently got a new camera to update my aging (gracefully!) Olympus E-P1. Deciding to stay with the µ⁴⁄_3^rds (or µ43^rds, or micro four-thirds - I just wanted to play with some HTML entity codes), I found a really tasty new Olympus OM-D E-M5 waiting for me for Christmas! In deciding which kit lens to go ahead and get with it, the Olympus M.ZUIKO DIGITAL ED 12-50mm F3.5-6.3 EZ was a no brainer. Water-resistant (to match the camera), and a neat macro function!

Well, I couldn't help but snap a couple of macro shots of my trusty Warhammer 40k Space Marine to play with this newfangled lens.

The problem, if you haven't done macro photography before, is that the depth of field will become razor-thin. You can alleviate it a bit by closing down the aperture a bit, but at some point you'll hit diffraction limits (and reduced sharpness).

Focus Stacking

So, what if you want a deeper depth of field in your shots? Well, there's a neat technique called "Focus Stacking" where you will take multiple images, with the only thing that varies between them is the focal point. Look again at the lead image, and specifically notice how the focus plane shifts along the length of the gun body.

Basically, compose and capture an image with the focus at one point, then focus a little further into your image, repeat as often as needed to have the range you want all in focus.

Focus stepping across my subject.

The problem, of course, is that you now have multiple images, all with different focus planes. How can you combine them all into one image where the different in-focus areas are all combined?

Well, we're going to use the open source Enfuse to do this dirty work for us. First, though, we need to make sure the images are properly aligned first...

Hugin Align Images

Well, Enfuse will work best if the images that are going to be combined are all as perfectly aligned as possible. To do this, we're going to use align_image_stack from Hugin.

Hugin is a really useful tool to have installed if you haven't already. To follow along with any of this, you'll need to install it anyway - so go do that now. Once installed, you can invoke the align_image_stack script from the bin/ directory of wherever you installed Hugin. It will look something like this (on Windows at the moment, so this is what the Windows command will look like on my machine):
C:\Program Files\Hugin\bin\align_image_stack -m -a OUT FILE1 FILE2 FILE3Don't Panic. To follow along, basically, put all of your focus stacked images into a single directory. Open a command prompt, and navigate to that directory.

Assuming your Hugin installed in the same place as mine, you can run the command as I show above. The -m switch will optimize the field of view for all images except the first. The -a OUT will prefix all the output files as OUT...tif, and FILE1 FILE2 etc... are the names of each of your focus stacked images you are starting with.

When it's done, you'll have a few new images in your directory named OUT000#.tif. These are your newly aligned output images, and will be what we work with from here on out.

Enter Enfuse

At this point you've got your images aligned perfectly (hopefully), and now all that is left is to mask each image so only the sharpest, in-focus portions show through. This is exactly what we are going to get Enfuse to do for us.

Enfuse can be run from the command line (the same bin/ directory where align_image_stacks was located) as part of a Hugin installation, or there is a GUI called EnfuseGUI that will expose all the options along with fast previews.

The EnfuseGUI interface.

I'll cover both here just to be thorough.

From the manual for Enfuse, there is a section that specifically addresses focus stacking. In particular, they refer to a bare-bones call for Enfuse to stack the images:

enfuse \

--exposure-weight=0 \

--saturation-weight=0 \

--contrast-weight=1 \

--hard-mask \

... \

--output=output.tif \

input-<0000-9999>.tif

The same options entered in EnfuseGUI are located at "Fusion options":

Base focus stacking parameters in EnfuseGUI.

Exposure and Saturation weighting should be set to 0, while Contrast weighting should be set to 1 and Force hard blend masks should be checked. These options will not be changed as you dial in your focus stacking parameters.

You can actually go ahead and run it at this point to see what your output will look like. It should run reasonably quick, and you can investigate the results. The nice thing about EnfuseGUI is that you can generate a fast preview of the results using the Preview button instead of a full run (this is very useful while you tweak parameters to see what type of effect they'll have).

EnfuseGUI preview of base focus stacking parameters.

If you're happy with these results, then stop here. That's it! Quick, relatively painless, and you're happy with the results. Go have a beer! (Seriously, many people will be fine with these results, and don't have to head any further).

If you're curious whats happening behind-the-scenes, you can pass the option:--save-masksto have Enfuse save each of the masks that it creates. Below, for example, are the masks that were generated for the above image. You can clearly see the in-focus plane in white in these masks:

Masks generated by Enfuse (white areas in-focus).

Enfuse Expert Options

If you're still here, I am going to assume that you are less than satisfied with the results.

If you pay careful attention to high-contrast edges in your image, you may notice some faint "ghosting" on either side. On the dark side of a high contrast edge, you may notice a faint, light seam, and vice-versa on the light side. There may also be some slight "smearing" in areas that obscure some details.

One of the options that I find to produce slightly better results are to use the L* channel of LAB colorspace as the grayscale projector to Enfuse (don't worry about the technical stuff - basically use the L* channel in LAB to create the masks as opposed to averaging).

Using L* from LAB to create masks

If you notice some haloing effects around high contrast edges, you can mitigate them a bit by increasing the size of the box Enfuse uses to evaluate contrast detection. The option to adjust this box size is:
--contrast-window-size=XWhere X ≥ 3. The documentation points out that values larger than 5 might produce blurry images, but I would experiment with values all the way up to 9 to see how it effects your final output.

There is another good option to tweak your final output, and that is to enable the Laplacian Edge Detection algorithm. Laplacian Edge Detection can be enabled by passing a positive value to:
--contrast-edge-scale=XWhere X typically ranges between 0.1 - 0.5 (pixels), and 0.3 is suggested as a good starting point.

Unfortunately, due to a bug in EnfuseGUI, the GUI won't accept values that are commonly used for the correct parameter. The GUI only accepts integer values, with no way to pass a decimal value (0.3 for instance).

No problem! We can just switch gears here at this point and fall back to the trusty command line. So at this point, besides the base parameters, I am going to add in contrast-edge-scale to enable the Laplacian Edge Detection algorithm:

enfuse \

--exposure-weight=0 \

--saturation-weight=0 \

--contrast-weight=1 \

--hard-mask \

--gray-projector=l-star \

--contrast-edge-scale=0.3 \

--output=output.tif \

input-<0000-9999>.tif

You can compare the results of adding in the l-star option and using the Laplacian edge detection at 0.3 pixels below:

Base stacking parameters detail (mouseover for contrast-edge-scale=0.3)

In Summary

Whew! That was a lot of information. Let me summarize the steps here. Assuming you have all of your stack images in the same directory:

Use Hugin's align_image_stack to make sure all of your images are perfectly aligned.align_image_stack -m -a OUT FILE1 FILE2 FILE3
Try Enfuse with the base focus stacking parameters to see if you are happy with the results.enfuse --exposure-weight=0 --saturation-weight=0 --contrast-weight=1 --hard-mask --output=base.tif OUT*.tif
You can also try a different type of grayscale conversion, I've found the L* channel from LAB to be pretty good overall ( --gray-projector=l-star ).enfuse --exposure-weight=0 --saturation-weight=0 --contrast-weight=1 --hard-mask --gray-projector=l-star --output=base.tif OUT*.tif
If there are halos around high contrast edges, you can try increasing the contrast-window-size a bit. ( --contrast-window-size=5 ).enfuse --exposure-weight=0 --saturation-weight=0 --contrast-weight=1 --hard-mask --contrast-window-size=5 --output=base.tif OUT*.tif
If details are smeared, try using Laplacian Edge Detection instead. (Adding --contrast-edge-scale=0.3 to the base command), and possibly removing --contrast-window-size.enfuse --exposure-weight=0 --saturation-weight=0 --contrast-weight=1 --hard-mask --contrast-edge-scale=0.3 --output=base.tif OUT*.tifModify the value between 0.1 and 0.5 to see which result is best for you.

Thoughts

If you want a truly sharp macro shot with a large, smooth depth of field, you'll be taking a lot of photos. Seriously. I've seen 100+ images used for a single macro shot. Look at these:

Red eyes by Gilles San Martin, on Flickr. 52 image stack.

Xylotrechus arvicola by Gilles San Martin, on Flickr. 55 image stack.

Try keeping an overlap between each of your shots. It'll avoid any seams in the final result. Remember, Garbage in - Garbage out.

The documentation offers a good tip as well - if the in-focus to out-of-focus transition is too sharp, try taking another image at the transition area with a larger aperture to allow a smoother, more natural falloff to occur.

Thats it! Hopefully this will be helpful to somebody, somewhere! If you found this walkthrough helpful in any way, please feel free to share it with others (or conisder donating 25¢ for my next cup of coffee by clicking here: - it helps to keep me awake cranking these tutorials out!)

Some References

The Enfuse documentation (section on Focus Stacking)
How to obtain a greater depth of field in macro photography - EDU PÉREZ
EnfuseGUI

Getting Around in GIMP - Black and White Conversion (Part 4)

2013-01-11T09:04:00.000-06:00

The first part of this tutorial looked at GIMP desaturate to convert to grayscale, the second part investigated using the Channel Mixer to decompose to grayscale with varying contributions from red, green, and blue, and the third part looked at decomposing an image into its constituent color channels in various modes.

This part of the tutorial will focus on a couple of semi-automated methods for converting to black and white, as well as utilizing GIMP layer blending modes.

Pseudogrey

I had previously written about achieving True Pseudogrey in GIMP, including writing a quick script-fu to automate the process.

The basic approach in Pseudogrey is that you can achieve a much higher density of tones converting to B&W by allowing single pixels to stray just a tiny bit away from pure gray (basically add an imperceptible amount of red, green, or blue in order to provide a finer gradation of luminosity values). I refer you to the original post for the full technical explanation.

The results from using Pseudogrey will follow the same model as for Luminosity desaturation, but will produce a larger range of luminosity tones (1786 possible shades vs. 256).

There are a couple of ways to convert your images to pseudogrey. You can use the script-fu I wrote previously by downloading it from one of the places below:

You can download the Script-Fu for Pseudogrey here:
Pseudogrey on GIMP Registry
or you can get it here:
Pseudogrey on Google Drive

Alternatively, if you've installed G'MIC (which you really, really should do), then you'll find it under the Black & white options. There will be an option for Pseudo-gray dithering at different levels.

If set to 1, it should yield the same type of result as my script-fu (technically, G'MIC pseudogrey dithering should be more accurate, though). Higher levels will allow stronger color casts to be applied to each pixel to increase the luminosity range even further.

GIMP desaturate (L) vs. Pseudogrey Level 1 (R). Notice the banding is mostly gone.

Pseudogrey is especially helpful in areas with very slight tonal value changes, as this is often where banding will become visible in an 8-bit image. In the example above, the left image shows banding that is mostly eliminated with just a single pseudogrey level on the right.

Sometimes the converted image will look good to you just as it is, but more often than not I will use a Pseudogrey conversion as a base layer under further manipulations. I'll just adjust the opacity of the layers above it to allow a bit of the Pseudogrey layer to show through (or a lot, depending on my mood/taste).

GEGL C2G

The Generic Graphics Library (GEGL) is the underlying graphics engine for GIMP going forward. There is one neat function in GEGL specifically for B&W conversions, and that is Color2Grayscale (c2g). It can be called from the Tools menu in GIMP:

Tools → GEGL Operation...

In episode 84 of Meet the GIMP, Rolf covers c2g briefly. Paul Bou also looks at using c2g for B&W conversions in a little more detail, and Joel Cornuz also asks if c2g could be the "ulimate" B&W converter.

I don't know if c2g is worth all of the hyperbole, but it does some very neat things.

What c2g will do is consider each pixel relative to it's neighbors within a given radius, and will adjust the value of the pixel to a gray. This gray value is evaluated as a function of perceived luminance weighted against it's neighbors.

The actual description of what c2g does from GEGL.org is:

Color to grayscale conversion, uses envelopes formed from spatial color differences to perform color-feature preserving grayscale spatial contrast enhancement.

In practice, what c2g will do is attempt to scale the values of pixels in it's neighborhood (radius) to contain black, white, and gray tones between. This will lead to a much more contrasty image than you started with.

What some people love about c2g is that it will also introduce a nice range of "grain" during it's conversion. If you don't care for this noise/grain, then it can be ameliorated by adjusting settings.

Straight desaturation in GIMP (luminosity).

At first glance, GEGL c2g is ugly. The default settings are not conducive to producing a pretty image. Have a look:

c2g conversion, default settings (radius 300, samples 4, iterations 10).

These default settings will (usually) produce a nasty halo effect on edges where the radius is not large enough to fully consider transitions. Look at the edges of the trees/buildings against the sky to see what I mean. There is also (to my taste) and excessive amount of synthetic graininess to the result.

If you've got some time (seriously, the conversion is very slow) you can achieve better results by tweaking the parameters.

You can decrease halo effects by increasing the radius over which the algorithm looks at colors. You can decrease graininess by increasing the samples a bit, or by increasing iterations as well. Iterations seem to have a larger effect on overall noisiness in the results, but at the cost of a longer processing time.

Better results after increasing some parameters (radius 750, samples 8, iterations 15).

This looks better. Increasing the radius helps to alleviate some of the halos, and will allow the algorithm to spread out the contrast over a larger area. The increase in samples and iterations also helped to keep the noise down to a more manageable level. At this level, some may prefer to keep the noise as a sort of "digital grain" in their image.

If you are willing to wait a bit more, you can refine the results even further:

Something I might use! (radius 1500, samples 8, iterations 20).

At this point, the noise is nicely suppressed while the halos have mostly been eliminated. The overall image is still has more contrast than a straight luminosity desaturation, but it's contrast that has been weighted for the surrounding pixels as well.

Basically, if a luminosity desaturation will choose a pixel value based on a color brightness approximation to your eye, c2g will choose the pixel value based on the value of surrounding pixels.

Perhaps you are familiar with this optical illusion showing the effect on perceived luminosity relative to nearby brightness?

Square A and B are the same shade of grey!

Squares A & B are the same exact shade of grey (open the image in GIMP and check for yourself if you don't believe me. :) ). The reason we perceive B as lighter than A is due to the way our eyes are perceiving nearby colors (and our expectations that are strengthened by the checkerboard pattern).

Running this image through c2g will align the pixel values closer with what our eye might expect to see:

After letting c2g do it's thing.

After running through c2g, if you check the value average over a few pixels in the squares, you'll now see that the values have been shifted so that, in fact, square B will be lighter than square A (also the overall contrast has been increased as well).

In short, c2g can be very handy for bringing out micro-contrasts in an image (or increasing global contrast at large radius settings). I rarely use c2g by itself anymore these days, and instead use the results to further refine my images if I feel I want more contrast in some areas (along with a layer mask).

Layer Blending Modes

I'm a little hesitant to cover this particular area of conversion, only because it is not one that I use normally in my workflow. Occasionally I will try this method out, but then usually revert to layering other layers with different results and masking as desired.

The best thing about an approach like this is that you can get instantaneous feedback on the results in your image directly on the canvas.

The basic idea is to use one of GIMPs layer blending modes to convert the image to grayscale, and then to modify an underlying color version of the image to adjust tones.

There's a few different ways to approach this:

Value

Open your image. Add an all black/white layer underneath it. Set the original (color) image to layer blend mode Value.

This will use the Value of your color layer, but the hue and saturation will be from the layer beneath it (all black/white). Basically, remove Hue/Saturation from your image, and only show its Value.

Now, you can adjust the colors of the original color layer, and effect what the final Value will be. You can adjust any of the color channel contributions like normal, and you'll be presented with an immediate view on your canvas of what the results will look like. You can use Levels, Curves, or Hue/Saturation/Lightness to make the adjustments. (Really any method that you like that will adjust the colors of the image).

Saturation

In this method, you'll add a black (or white) layer above your base color layer, and set the layer mode of the black/white layer to Saturation.

This will use the Saturation of the upper layer, and the Hue/Value from the lower (color) layer.

As before, you can now proceed to adjust the colors of the color layer to taste, and you will see the results of your modifications directly on your canvas.

Color

This is similar to the Saturation approach above, but in this case you'll set the black/white layer to layer mode Color.

This will use the Color from the black/white layer, and the Hue/Saturation from the layer below it (color).

Now adjust colors of the color layer to taste.

* Amelia Bellamy-Royds notes in this Flickr thread that the Color option appears to only operate on the Red channel, but I haven't fully tested this yet to see what is going on.

Hue/Saturation

You can also achieve similar results directly from a color layer by invoking the Hue/Saturation dialog.

Just set the Master saturation to 0, then you can adjust each of the RGB/CMY channels separately to modify the results (and still get live updates in your canvas on screen).

Play With It

The key to the layer blending modes (and Hue/Sat adjustments) is to experiment a bit to get accustomed to what your changes are affecting. Start with single channel adjustments and a goal in mind. If you want to darken up the blues in the sky, play with that channel a bit in one of these blending modes and see how it affects things.

You'll have a similar level of control that you would achieve with using Channel Mixer, but with the added benefit of having a full-canvas preview of the effect (as opposed to the small preview window in Channel Mixer).

Nearing the End

I think I'm running out of gas here, but the good news is that the end of the tunnel is in sight (or a train is speeding my way). Either way, the end is near (the next time I attempt tackling a subject like this, please remind me about publishing this series...). If you've followed me this far, you deserve a medal, or a certificate or something!

Stay Tuned

In the last installment in this series, I'll have a look at combining everything we've seen so far, and to look at ways to blend them together. This is more often my general workflow these days.

← Part 3 (Decompose)Part 5 (The End) →

(Abandoned) Splendid China Florida

2013-01-07T14:53:00.000-06:00

I was fortunate enough on a recent family trip to stay in a vacation home in Kissimmee, FL for a week. We had passes to Disney, and made good use of them. With kids, though, late nights were not going to be possible.

Luckily for me, on the drive to the house itself, there was a rather large property off the road that was surrounded by a wall that was just barely containing some wild overgrowth. Obviously an abandoned property, it piqued my interest enough to have me take a quick gander at Google maps to see what lay beyond the wall.

Imagine my surprise when I realized that just a few blocks from where I was staying was the rotting carcass of the Splendid China theme park!

The Splendid China themepark map.

This was too good of an opportunity to pass up (I am drawn to shooting abandoned places, but lack a good selection to choose from where I live). Some of my favorite photography is from abandoned places (thank you Opacity.us!), so I had to give it a try.

Splendid China - Potala Palace by Pat David on Flickr.

It was about 8:30 PM before I finally made it inside, and found myself on the back side of the Potala Palace (seen above). It was cold (around 40° F), and I was lucky enough to have a full moon to light my way. The area is not fully developed around the park, so it was extra quiet and still in the night.

The paths were still relatively clear of undergrowth, but I still had to be careful when walking around (there were some small "lakes" and other water features scattered around, so keeping to the path seemed like a good idea).

Monument by Pat David on Flickr.

The monument above actually had a small moat surrounding it (just past the underbrush in the foreground). I had a chance to stroll through the paths in the bright moonlight, and I had a particular destination I wanted to make sure I didn't miss. The Imperial Palace/Forbidden City!

Meridian Gate - Forbidden City by Pat David on Flickr.

Everything in the park is ¹ ⁄ _10^th scale, and was handbuilt by artisans imported from China for the job. What's amazing is that in some of these photos you can clearly see that the roof tiles, for instance, were all tiny, individual, actual tiles scaled down! The craftsmanship was truly remarkable here.

Meridian Gate detail - Forbidden City by Pat David on Flickr.

Despite the best efforts of vandals, the buildings were in remarkably good shape. There was some graffiti, and some vanadalism to the structures, but overall they seem to have held up well after almost a decade of neglect!

Hall of Supreme Harmony - Forbidden City by Pat David on Flickr.

To put the scale in perspective, look at the image below. You'll find a standard electrical outlet box on the ground to the right of the staircase.

Hall of Supreme Harmony - Forbidden City by Pat David on Flickr.

Forbidden City by Pat David on Flickr.

Overall it was a fantastic evening spent strolling through the grounds of this abandoned park. I stayed mostly to the edges and back parts of the property as I didn't want to alert anyone to my presence. I also tried to stay as respectful as possible and to not damage anything. Just being there and able to shoot some photographs was more than enough for me.

I'll leave you with a few more images. Feel free to check out the rest of the set on Flickr!

Splendid China - Wooden Pagoda by Pat David on Flickr.

Splendid China - White Pagoda by Pat David on Flickr.

The white pagoda had a lot of graffiti on it, and the trees mostly obscured it from view.

The Great Wall by Pat David on Flickr.

The top portion of the Great Wall was actually built from millions (no exaggeration) of tiny, scaled down bricks!

Splendid China - Pagoda by Pat David on Flickr.

Most of these images were shot in the 30+ second range for exposure, and I still under-exposed by a stop or two! What this tells me is that I need faster lenses...

Getting Around in GIMP - Black and White Conversion (Part 3)

2012-12-07T14:57:00.001-06:00

In the first part of this tutorial we had a look at using the Desaturate command to convert images to grayscale. The second part of this tutorial examined the use of Channel Mixer to adjust the contributions of each Red, Green, and Blue channel to the final grayscale result. This part of the tutorial will focus on decomposing the entire image to its component parts for (possibly) further manipulations.

The rest of my GIMP tutorials can be found here:
Getting Around in GIMP
The rest of the tutorials in this series are here:
B&W Conversion - Part 1 (Desaturate)
B&W Conversion - Part 2 (Channel Mixer)
B&W Conversion - Part 4 (Pseudogrey/c2g/Layers)
B&W Conversion - Part 5 (Putting it All Together)

To get a good grasp of what we are about to do, it helps to remember the very first part of this tutorial when we looked at what goes into producing a color pixel on your screen (you remember the R, G, B sub-pixels, right?).

I've also written a small Script-Fu helper script to decompose an image layer to all the different color mode decompositions I'm listing here. Find it at the end of the page!

Decomposing Colors

Previously, in the Channel Mixer we saw that we can adjust the contribution of each of the RGB channels to our final grayscale conversion. From that post, you should have seen that you can isolate a specific channel by setting its value to 100, and leaving the others at 0.

Chad White by Ryan Abel on Flickr (CC-BY)

If you wanted each of the RGB channel contributions isolated into its own layer, it would be tedious to do it manually for each channel. Luckily for us, GIMP has a built in command to automatically Decompose your image into different channels:

Colors → Components → Decompose...

This command will let you decompose your image into each of its constituent channels.

The Decompose dialog.

The options you have to choose from include the Color model to decompose to, and whether to create a new image with the decomposed channels on layers. If you don't check Decompose to layers you'll end up with a new image for each channel (chances are you'll want to start out leaving this checked).

The most important option is which Color model you'll want to decompose to. Up to now we've mostly been focusing on RGB, but there are other modes that might be handy as well. Briefly, the modes most useful might be:

RGB(A)

This is the Color mode that we've been focusing on mostly up to now, and is usually the most helpful in terms of having multiple sources to draw from. This will separate out Red, Green, and Blue Channels into layers for you (and Alpha if your image has it).

RGB decomposed (in order): Red, Green, and Blue

HSV/HSL

Hue, Saturation, and Value/Lightness is another useful decomposition, though usually only for the Value or Lightness for our purposes.

Hue, Saturation, Value (HSV) Channels

The Value in HSV is derived according to the formula:

Value, V = MAX(R,G,B)

Which is basically just the largest value of R,G, or B.

Hue, Saturation, Lightness (HSL) Channels

The Lightness in HSL is derived according to this formula:

Lightness, L = ^{[ MAX(R,G,B) + MIN(R,G,B) ]} ⁄ ₂

Lightness is really just the average of the largest and smallest component of RGB.

While Hue and Saturation may seem interesting, it should be fairly obvious that the channels most useful for what we want to do would likely be Value or Lightness. Overall Lightness will tend to be a bit brighter than Value.

LAB

There is far too much information concerning the LAB colorspace to really go into much detail here. Suffice it to say that the L in LAB is for Lightness, while A and B are for color opponents (A = Green↔Red, B=Blue↔Yellow).

When I get out of the B&W world later, there are some neat tricks you can do in LAB colorspace for adjusting the color toning of your images I'll try to cover.

The LAB colorspace is based on a perceptual model (similar to the relative luminance from the last post). In fact, the Lightness in LAB is calculated using the cube root of the luminance from that function.

LAB Channels

As you can see, the only channel of any use for our B&W conversion is really the Lightness, L channel.

CMY(K)

Cyan, Magenta, Yellow and (K, Black) is often discussed in terms of printing, but can be useful in digital manipulations as well. In particular, the CMY conversion will produce three channels of Cyan, Magenta, and Yellow. When you do the decomposition in GIMP, you'll need to invert the channel colors to really inspect what they produce (Colors → Invert).

CMY conversion (inverted from direct conversion)

CMYK produces a similar result, but adds another channel to control the level of black in the result. As with the CMY decomposition, you'll need to invert the Black, K channel.

CMYK conversion, with the Black, K channel inverted.

YCbCr

Anyone who does video processing might recognize this colorspace representation, as it shows up often in digital video. YCbCr is a means for encoding the RGB colorspace, and produces three channels, a Luma, Y and two channels of Red (Cr) and Blue (Cb) chroma differences.

YCbCr (ITU R470 256) decomposition.

If you use this decomposition, try to use the 256 variants of the ITU recommendations to allow the decomposition to utilize the full 256 values available (the non-256 versions will pad 16 to the range - only allowing values to go from 16-240).

So Now What?

Having looked at this, let's summarize some of the most useful Colors → Components → Decompose results:

RGB - All
HSV/HSL - V(Value) and L(Lightness)
LAB - L
CMY - All
CMYK - K
YCbCr - Y(Luma)

This gives us a total of 12 different types of color mode conversions that may be useful to us in generating a B&W image. It helps to visually be able to see all of the options at once to get a better feel for what is going on:

Chances are that one of these conversions might prove useful to you as a direct B&W conversion. If so - perfect!

Personally I am gravitating towards the Red channel, or the Cyan channel if I had to choose one. I like having the texture and tones of the shirt be a little darker to differentiate it from the pants. This is a very subjective call, of course. :)

It helps to notice that the first 6 conversions above are all color channels (RGB and CMY). This means that they will vary from image to image depending on dominant colors.

The last 6 conversions are all brightness values based on different functions for achieving the result (blacK, Value, Lightness, L, Y (luma)).

Looking Forward

Chances are better that some parts of some conversions will be useful. I personally am rarely satisfied with any of the straight conversion options on their own, but would like to pick and choose which parts of the image contain the best detail and tones from different conversion options (this includes any of the procedures we've seen up to now, by the way). Then I combine them as needed to fill out the tonal details in my final result.

Combining these methods that we've seen so far is the basis for the last part of this tutorial, so stay tuned!

The Script

While writing this post, I had to decompose our base image into each of the different channels to create images. It was tedious, having to choose each color model manually, and decomposing to another image with layers.

So I wrote a script to automate this process for me. It will basically take the active layer in an image, and decompose it to each of the useful color channels listed above, each as its own layer. (It will also invert some of the layers that need it).

Once you've placed it in your Scripts folder, the command can be found here:

Colors → Color Decompose...

You can find your Scripts folder by going to:

Edit → Preferences

Then from the Preferences dialog, down to:

Folders → Scripts

You can download the Script-Fu for Color Decompose here:
Color Decompose on GIMP Registry
or you can download it here:
Color Decompose on Google Drive

Stay Tuned

We've got Pseudogrey and Layer Modes coming up for other options to get you to your perfect B&W conversion! A little later we'll see how to combine all the results we've seen so far into our final image for manipulation.

← Part 2 (Channel Mixer)Part 4 (Layers/c2g) →

Getting Around in GIMP - Black and White Conversion (Part 2)

2012-11-20T17:41:00.002-06:00

In the first part of this tutorial we had a look at using the Desaturate command to convert images to grayscale, and how the different options in that command work to produce their results.

The rest of my GIMP tutorials can be found here:
Getting Around in GIMP
The rest of the tutorials in this series are here:
B&W Conversion - Part 1 (Desaturate)
B&W Conversion - Part 3 (Decompose)
B&W Conversion - Part 4 (Pseudogrey/c2g/Layers)
B&W Conversion - Part 5 (Putting it All Together)

We saw how the Desaturate command can use straight numerical evaluations for conversion (Lightness and Average) as well as using the relative luminosity model for how our eyes will perceive brightness based on color.

These are fantastic and easy to use conversion options that require no extra work on your part to get to a grayscale image. Almost every other option from here on out will require you to make choices and adjustments to get your results.

This time around we are going to have a look at another very widely used method for converting to grayscale, the Channel Mixer.

Channel Mixer

Using Desaturate let you convert to grayscale based on pre-defined functions for calculating the final value, but what if you wanted even further control? What if you wanted to decide just how much the red channel should influence the final gray value, or to have more control over the ratios and weightings of each of the different channels independently? That is precisely what the Channel Mixer will allow you to do.

For the examples below, I'm going to mix it up with a different color gradient test map, blue to blue HSV gradient, with a gradient to black vertically. This represents our entire colorspace (feel free to download this image to follow along).

Gradient representing all colors and shades in our colorspace.
To compare using Desaturate: Original - Lightness - Average - Luminosity

So let's have a look. The Channel Mixer can be invoked from:

Colors → Components → Channel Mixer...

When it opens you'll be presented with the main window:

Now, you can use the Channel Mixer in your full color mode if you'd like, but I'm not going to go into that here. We want to get a grayscale conversion, so you can check the option for Monochrome (this will grey out the Output channel option in the dialog). This will turn your preview into a grayscale image.

I'll come back to the Preserve Luminosity option shortly.

Warning: Math Ahead

If all you did was check the Monochrome option, and if you left the Red channel slider at 100, then you'll be seeing a representation of your image with no Green or Blue contribution (ie: you'll basically be seeing the Red channel for your image):

Basically just the Red Channel

What this means is that with Green and Blue set to 0, it will look at the value of the Red in your pixel, and map it directly to a grayscale value. If your pixel RGB values were: 200, 150, 100, then the Value for the pixel would become 200, 200, 200.

The sliders represent a percent contribution to the final value.

For instance, if you set the Red and Green channels to 50(%), you should see something like this:

Red at 50%, Green at 50%, Blue at 0%

In this case, Red and Green will contribute 50% of their values (and nothing from Blue) to the final pixel gray value. Considering the same pixel from above: 200, 150, 100, we would get:

(200 × 0.50) + (150 × 0.50) + (100 × 0)
(100) + (75) + (0) = 175

So the final grayscale pixel value would be set to: 175, 175, 175.

Preserve Luminosity

"These go up to 11." - Nigel Tufnel

The astute will notice that the sliders actually have a range from -200 to 200. So you may be asking - what happens if two channels contribute more than what is possible to show?

For instance, what if both the Red and Green channels were set to contribute 100%?

(200 × 1.00) + (150 × 1.00) + (100 × 0) = 350

Well, Channel Mixer will allow us to set these values, but we can't very well set the grayscale pixel value to be 350. So anything above 255 will simply end up being clipped to 255 (effectively throwing away any tones above 255 - bad!).

This means that you have to be careful to make sure that each of the three channels contributions don't exceed 100 between all of them. 50% Red, 50% Green is ok - but 50% Red, 50% Green, AND 50% Blue (150%) will clip your data!

This is where the Preserve Luminosity option comes into play. This option will scale your final values so the effective result will always add up to 100%. So, the scale factor from above would be calculated as:

¹⁄_{(1.00 + 1.00 + 0)} = 0.5

So in our case above, the 350 would be scaled in half (0.5), giving the actual final value as 175. If Preserve Luminosity is checked, all the values will be scaled by this amount. (No clipping = good).

This is not to say that you need to always use Preserve Luminosity, just stay aware of the possible effects if you don't.

Speaking of Luminosity

You'll notice that Channel Mixer basically gives you control over each channels weightings, and in the previous post I showed the function used for desaturating according to relative luminance. If you wanted to replicate the same results that Desaturate → Luminosity produce, you can just set the RGB sliders to the same values from that function (21.3, 71.5, 7.2).

Replicating the luminosity function.

If you're just getting started with Channel Mixer, this makes a pretty nice starting point for you to begin fiddling.

Begin Fiddling

It's tough to find representative images to demonstrate using Channel Mixer, but I managed to find a very pretty landscape by Cyndi Calhoun on Flickr called Garden of the Gods - Looking North (cc-by):

Garden of the Gods - Looking North

If possible, you'll want to keep in mind the primary RGB influences in different portions of your image as a means for approaching your adjustments. For instance, this image (not coincidentally) happens to have strong Red features (the rocks), Blue features (the sky), and Green features (the trees).

I am going to want to try to keep any of the individual channels from getting so bright you lose detail (blowouts), or from crushing shadows too much. Remember, you want to try and keep as much tonal detail as possible!

So I'll use the luminosity function as a starting point...

Straight conversion using the luminosity function.

It's not a bad start at all, but I feel like the prominence of the red rocks in the sunlight has been dulled quite a bit. It's a central feature of the image, and should really draw the eye towards it. So we need more pronounced reds to make the stone pop a little bit more.

I'll turn on Preserve Luminosity, and begin bumping the Red channel to taste.

Red channel bumped up to 66.1

This gives a little more prominence to the red stone, I think.

The Green channel feels like it might be ok to me, but for comparison, try lowering it to about half of the Red channel value. (Remember - Preserve Luminosity is checked - so the final values will scale to give Red values twice the weight as Green now).

Green channel value at ~half of Red.

If you're wondering why the red rocks got brighter, consider the math. Previously, Red and Green were very near each other in value (around 70), so both colors had approximately equal weight (There was more red than green in the rocks, obviously). When Green got its influence cut in half, red scaled to take a much larger influence, and because there was more red than green, the final value will end up higher.

So, if we look at the RGB values of the rock, we'll find that it's roughly like this (we're ignoring Blue at the moment because for this example it's staying constant):

226, 127

If both Red and Green have equal weighting, the final pixel value will be:

(226 × 0.5) + (127 × 0.5) = 176.5

Now if Green is only half as strong as Red, the value will be:

^{((226 × 0.5) + (127 × 0.25))} ⁄ _{(0.5 + 0.25)} = 193

I had to divide the result by the influence amount to scale the way Preserve Luminosity would. You can see that the final pixel value will become brighter in this case. This is why the red rocks get brighter when you decrease the Green channel.

It should go without saying, but the Blue channel will have a lot of influence on the sky (and many areas of the image in shadow). So if we wanted the sky to look a little more dramatic, we could remove the Blue channel influence by setting it to 0:

Same as before, but Blue channel set to 0

This will darken the sky up a bit (as well as some shadow areas).

You'll need to pay attention to what these changes do to your image in closer views sometimes, though. In our case, there is a higher amount of banding and noise in the smooth sky if values get pushed to extremes. So try to approach this with a light hand if possible.

The sliders also allow negative values. This will seriously crush the channel results when applied (and will quickly lead to funky results if you're not careful). For example, to push the Blue channel even darker in the final result, I've set the Blue channel to -20 here:

Red: 66.1, Green: 33, Blue: -20

The sky has become much darker, as have the shadow sides of the rocks. There is an overall increase in contrast here as well, but at the expense of nasty noise and banding artifacts - especially in the sky.

Rules of Thumb
Generally, the Red channel is well suited for contrast (especially in the brighter tones). The Green channel will hold most of the details, and the Blue channel contains grain and (usually) a lot of noise.

In skin, the Red channel is very flattering to the final result, and you'll often get great results by emphasizing the Red channel in portraits.

On Skin

The Red channel can be very flattering on skin, and is a great tool to keep in mind when producing portraits. For instance, below is the color image of Whitney from the first part of this tutorial:

Whitney in color.

The straight luminosity conversion is below. Mouseover the image to compare it to one where the Red channel is set equal to the Green channel (giving greater emphasis on Reds):

Whitney in Luminosity (mouseover to see Red channel = Green channel).

B&W Film Simulation

Due to the popularity of the Channel Mixer as a straightforward means of conversion with nice control over each of the RGB channel contributions, many people have used it as a basis for building profiles of what they felt was a close emulation to classic black and white films.

I've borrowed the table from Petteri Sulonen's site that shows some commonly used RGB Channel Mixer values to emulate B&W films. These aren't exact, of course, but some of you may find them useful, so I wanted to include them here.

Agfa 200X	18,41,41
Agfapan 25	25,39,36
Agfapan 100	21,40,39
Agfapan 400	20,41,39
Ilford Delta 100	21,42,37
Ilford Delta 400	22,42,36
Ilford Delta 400 Pro & 3200	31,36,33
Ilford FP4	28,41,31
Ilford HP5	23,37,40
Ilford Pan F	33,36,31
Ilford SFX	36,31,33
Ilford XP2 Super	21,42,37
Kodak Tmax 100	24,37,39
Kodak Tmax 400	27,36,37
Kodak Tri-X	25,35,40

Bonus!

I've written and added a Script-Fu to registry.gimp.org that will automatically create new layers for each type of film emulation in the table above, with the proper channel mixer settings. You can also choose to create a "contact sheet" of the converted layers for comparison instead.

Download the B&W Channel Mixer Script-Fu at the registry!

In Summary

There's a good reason that the Channel Mixer is such a popular means for converting an image to grayscale. It's flexible, and allows a great level of control over the contributions from each channel that makes up your final image.

Unfortunately, the only way to preview what you are doing is in the tiny dialog window. Even with zooming it can sometimes be frustrating to make fine adjustments to the channel contributions.

I encourage you to give it a try on some of your own photos to see the effects that are produced by emphasizing some channels over others. This will become important later as we look at using entire decomposed channels of data, and ways to blend and mix them to produce our final result.

Stay Tuned

In a few days I'll be back again to have a look at a slightly more intensive way of approaching your grayscale conversions: Decomposition (RGB, LAB, etc).

Part 3 will show how we can decompose an image and use the different channel layers to build up to a grayscale image! It's a slightly more intensive process, but the level of control is fantastic and beyond anything we've talked about so far.

← Part 1 (Desaturate)Part 3 (Decompose) →

Getting Around in GIMP - Black and White Conversion (Part 1)

2012-11-14T11:10:00.000-06:00

This is a long topic, so to keep you from wanting to put your eyes out with a spoon, I've tried to break things up a bit. In this first part, I'll look at using the GIMP Desaturate command to reduce your images to grayscale and to hopefully shed some light on just how the options calculate exactly what level of gray each pixel should be.

The rest of my GIMP tutorials can be found here:
Getting Around in GIMP
The rest of the tutorials in this series are here:
B&W Conversion - Part 2 (Channel Mixer)
B&W Conversion - Part 3 (Decompose)
B&W Conversion - Part 4 (Pseudogrey/c2g/Layers)
B&W Conversion - Part 5 (Putting it All Together)

What We're Trying to Achieve

B&W photography deserves a much longer look than I can afford to bore you with here. However, there are a few things I would like to focus on in regards to preparing your images for B&W.

What you want to keep in mind is that by removing the color information you have effectively left yourself with only tonal data (and composition) to convey your intentions.

This can be both liberating, and confining.

By liberating yourself of color data, you can focus much more clearly on the subjects and composition with whats left. (Indeed, this is often felt to be one of the primary reasons street photography is normally associated with B&W images - with no colors to distract you, the focus is on the subjects and composition even more).

Without color, the form and tones are all that's left. (Edward Weston, Pepper #30)

Tonality

What I tend to refer to when using this term is the presence and relationships between different values of gray in the image. This can be subtle, with smooth, even differences among values, or much more pronounced.

Contrast

Contrast is the relative different in brightness between parts of an image. High contrast will have a much sharper differentiation between lighter and darker portions of an image, while low contrast will show less differences. Often a straight conversion to B&W can result in gray values that are all very similar, yielding a visually "flat" image.

Globally, it may refer to the overall distribution of lights and darks. Locally the same definition applies, but applied across a smaller section of the image.

Dynamic Range

Dynamic range is just the overall range of values being captured in your image. It represents the maximum dark and light that is captured for a given exposure. (Extending this dynamic range in an image is the topic for a later tutorial).

The Approach

The approach I will take here follows similar approaches I had taken in film days. I'll attempt to use different methods of grayscale conversion (and blending them) to get to a working image that is as full of tonal detail overall as possible. (Petteri Sulonen refers to this as his "digital negative") - if you want a great look at a digital B&W workflow, head over and read his article.

Then with the image containing as much tonal detail as possible, I will approach it with adjustments of various types to produce a final result that pleases my eye.

Before we can head down that road, we have to step back and consider the tools we are using. So...

Let's have a look at how an image gets displayed on your monitor!

Your Pixels and You

We are working in an RGB world when we stare at our monitors. Every single pixel is composed of 3 sub pixels of Red, Green, and Blue.

300X magnification of an LCD panel. (Image from wikipedia)

Now, variations in the brightness of each of the sub-pixels will "mix" to produce the colors we finally see. The scales available to us (in an 8-bit display), are levels from 0-255 for each color (2⁸ = 256). So, if all the sub-pixel values are 0, you get black. If they are all 255, you'll see white. Any other combination will produce some variation on a color.

70, 203, 74 for instance

or 203, 70, 198

But what about 16-bit images?
Well - the data is still in the image file to correctly describe the colors at 16bit/channel, but most likely what you'll be seeing on your monitor is an interpolation of the values to an 8-bit/channel colorspace. You should always work in the highest bit depth color that you can, and leave any conversions to 8-bit for when you are saving your work to be viewed on a monitor only.

The thing to take away from this is to realize that when all three color channels are the same value, you'll get a gray color. So middle gray of 127, 127, 127 would look like this:

127, 127, 127

While this is a little brighter: 225, 225, 225

Very quickly you should realize that a true monochromatic grayscale image can show you up to 256 discrete shades of gray (And for 16-bit, 2¹⁶ will yield 65,536 different shades) going from 0 (pure black) to 255 (pure white). It is this limitation for purely gray 8-bit images that introduces artifacts in some images over smooth gradations (posterization or banding) - and is a good reason to keep your bit depths as high as possible.

In summary, realize that for a purely grayscale image, on an 8-bit monitor, you'll have 256 (visible) shades of gray to work with. (I say "visible" because if you're working in higher bit depths, the file will actually have many, many more shades of gray, but you'll still only be seeing 256 shades on your monitor).

Getting to Gray

There are many different paths to get to a grayscale image, and almost none of them are equal. They will all produce an image based on their different methods, and it will be up to you to decide which ones (or portions of) to keep and build on.

A combination of luminosity desaturation and GEGL Color2Grey (C2G).
12th Annual Dauphin St. Beer Fest - Conversation in Hayleys

For this tutorial I'm going to try to cover as many different methods as possible. This means we'll be having a look at:

Desaturate Command (Lightness, Luminosity, Average)
Channel Mixer
Decompose (RGB, LAB)
Pseudogrey
Layer Blending Modes
Combining of these methods

One of these methods may work fine for you. Or, if you're at all like me, it will most likely be a combination of one or more of these methods, blended through a combination of masking and opacity adjustments.

So in the first part of this tutorial we'll have a look at possibly the simplest and most straightforward method of converting your images to grayscale: Desaturate.

GIMP Desaturate

Perhaps the easiest and most straightforward path to a grayscale image is the Desaturate command. The command can be invoked from the Colors menu:

Colors → Desaturate

There are three options available from this menu:

Each of these options will generate a grayscale image for you, but the difference lies in the way they interperet your image colors into values of gray.

Base RGB gradient of pure colors.

Base RGB (additive color) mix.

Lightness

Lightness will add the largest value of R, G, or B, and the smallest value, and divide the result by 2.

¹⁄₂ × ( MAX(R,G,B) + MIN(R,G,B) )

Straight conversion yields similar values, regardless of color.

(Mouseover for color)

This means that (usually) one channel is actually ignored in the final result. The result will be a straight value conversion regardless of the color in consideration.

Average

Average will use the numerical average of the values of each pixel. (Again, regardless of the color in consideration).

¹⁄₃ × ( R + G + B )

Averaging, the values will tend darker overall.

(Mouseover for color)

Luminosity

Luminosity is the most interesting of the straight desaturating methods. The reason lies in the way that our eyes perceive brightness in colors.

Lightness and Average both evaluate the final value of gray as a purely mathematical function, without regard to the actual colors being used. Luminosity, on the other hand, utilizes the fact that our eyes will perceive green as lighter than red, and both as lighter than blues (relative luminance). This is also why your camera sensor (usually) has twice as many green detectors as red and blue.

(0.2126 × R) + (0.7152 × G) + (0.0722 × B)

This is closer to how our eyes will actually perceive brightness of the colors.

Notice the overwhelming contribution from green. (Mouseover for color)

I am not saying that any one of these is necessarily better for your conversion efforts. It depends on the desired results. However, if you are in doubt about which one to use – this is the best of the three options (to closest emulate what your eye will actually see).

A Couple of Examples

The image below, Joseph N. Langan Park, is an interesting example to see just how much green influences the conversion result in luminosity. Hover over the different conversion types in the caption to see them, and pay careful attention to what Luminosity does with the green bushes along the waters edge.

Mouseover type to see: Original - Lightness - Average - Luminosity

This shot of Whitney shows the effect on skin tones, as well as the change in her shirt color due to the heavy reds present. In just a Lightness conversion, the red shirt becomes relatively flat compared to her skin tones, but becomes darker and more pronounced using Luminosity. Her lips get a bit of a boost in tone in the Luminosity conversion as well.

Mouseover type to see: Original - Lightness - Average - Luminosity

Summary

There is no single right way to approach a B&W conversion overall, but hopefully this will give you some insights into how the GIMP Desaturate commands will view the data in your image.

Remember, the goal is to use these tools to help you obtain an image that has the highest tonal density possible with your source. This image will then be the base for further adjustments of contrast and levels to achieve your final vision. You may find that different portions of your image respond better to different conversion techniques (and that's fine - I'll cover that in the last part of this tutorial when we begin mixing these results together).

Stay Tuned

In a few days I'll be back for Part 2 to tackle another method for generating grayscale images - the Channel Mixer. It's one of the more popular methods I have seen used by others that offers a lot of power and flexibility in the process.

On to Part 2 (Channel Mixer) →

GIMP Editing Challenge Results

2012-10-31T14:50:00.000-05:00

At the beginning of this month I posted a RAW image of a house that I asked people to process in GIMP to turn it into a black and white image, and to describe their workflows in doing so.

My hope was to be able to expose some new and interesting ways to approach black and white processing in GIMP, and to give everyone a single point of reference to compare the results to.

The challenge base image

I had originally envisioned it as a sort of contest where everyone could vote on their favorites at the end, but it became cumbersome to maintain entries across Flickr and gimpchat.com. So instead I've left the posts as they are to hopefully help others to get a look into some cool ways to approach B&W conversions.

I was thinking that this might be a good image to test against because there is bright, just about blown, sunlight on the side of the house, while there are also very dark shadows in some of the windows and bushes. Hopefully a nicely challenging range of luminance and detail to work with!

There were some great responses from everyone who took a stab at it (pun intended - it is Halloween here!). I just wanted to take a moment to highlight a few of the images/processes that I liked personally...

Amelia Bellamy-Royds (AmeliaBr on Flickr)

Amelia did an amazing job, and was much more in line with what I personally was striving for - that is a well processed B&W image with good contrast overall. The shadows didn't fall into inky blackness, and the highs were not overexposed (beyond what was already blown in the source image).

She describes her methodology and workflow in this comment on the Flickr page.

Basically, she either applied a pure black (or white) layer under the color original, then set the blending mode of the original layer to "Value", or she desaturated the original, laying it over the original color layer - and then set the blend mode of the desaturated layer to "Saturation". In both cases she was extracting the Value of the pixel from HSV mode.

This is nice because it then allows her to adjust the levels/curves of the original color layer to affect the final monochrome result. This means she could then independently modify the Red, Green, or Blue curves to adjust their contribution to the final Value result. A fantastic and intuitive way of adjusting the results with instant feedback on the canvas!

Sebastian (absolute_rookie on Flickr)

Sebastian did things a little differently, and I love the result. There's a lovely feeling of softness in his, and the whole thing felt quite dreamy and is very pleasing to look at. He also details his process in the Flickr thread.

In summary, he opted to expose the RAW image three times (proper, under, over), and to merge the results back together using Luminance HDR (no word on which tonemapping operator he decided to use - Mantiuk maybe?). He then brought the results into GIMP and desaturated there (trial and error).

Starting from the RAW file, and under/over exposing a couple of versions likely helped retrieve some of the blownout highlights and shadow detail - so a great idea to approach it in this way!

Oh, and he also produced a really, really nice color version originally before he generated the black and white. You should check it out!

These were just a couple of the great images that resulted - there were many more, and I certainly don't want to exclude anyone. I recommend checking out the threads at Flickr and gimpchat to see some other results!

How did mine turn out?

Oh, and if anyone is curious how my own results came out, here is the edit I did the same day I first posted the challenge:

I detailed my process in the Flickr thread as well.

Basically, I started with a Pseudogrey base, then decomposed the original image to RGB layers, and brought each one as a layer separately into my workspace. Then I just used a combination of layer masks and opacity adjustments to tweak the contribution from each channel. (I did a lot of masking/burning/dodging around the bushes in the foreground - they just came out way too flat for me, and I wanted good separation between them and the mid ground).

This has been a really interesting experiment, and I've been blown away by the results and responses. I may just have to look around for another RAW image to post and see what everyone does with it as well.

Also, as I said in my first post on this challenge, I have a great big Black and White Conversion tutorial coming soon that will hopefully address all the different ways and methods for approaching a conversion. Stay Tuned!

GIMP Editing Challenge

2012-10-02T11:26:00.000-05:00

Over in the GIMP users group on Flickr, there has been a recurring challenge where a theme is chosen, and members will post images based around that theme. Then the users will all vote and choose their favorite, and the winner gets to choose a new theme. It's usually quite fun, and you get to see some really neat photos from other users.

The Challenge - Convert this to B&W

It occurred to me the other day, though, that it might be even more helpful to those learning to use GIMP to see better what other users workflows might be, and even better to see what the results might look like from different processes.

So I posted a simple challenge that is a little different from the usual one based around a theme. I posted one of my own images, and asked everyone to use the same base image to process according to a different theme (in my case I was curious about B&W conversion workflows).

I also asked that anyone entering the challenge to also document their workflow so others could hopefully learn some new things about the process that is used.

Not everyone that reads the posts here might be on Flickr, so I am posting this here to invite anyone else that wants to try out. (I am figuring this can run for a few weeks - maybe end on Halloween?)

The discussion page on Flickr is here:
Flickr.com: Challenge in GIMP - A Couple of Thoughts

And you can download the full size RAW file (or PNG) here:
Download .ORF RAW file (~12.5 MB)
Download the 16bit TIF (~69 MB)
Download the PNG file (~17 MB)

If you think you'd like to participate, it might be easiest to just join Flickr, and post your results in that thread (this way you can also vote on your favorite as well). If you don't want to join Flickr for some reason, you can send your final results to me, and I'll post on your behalf. (Just don't forget to include a writeup about the steps you took to produce your image).

I'm hoping this will be fun, and informative. It's also a sort-of prelude to a tutorial I'm in the process of writing about B&W conversion in GIMP. I'm hoping to have it out before/around Halloween.

Getting Around in GIMP - Heal Selection (Resynthesizer)

2012-08-27T10:31:00.000-05:00

Missing Tower! (from Eiffel Tower by Terrazzo on Flickr)

I've been reading through a ton of hyperbole about Adobes Photoshop "Content Aware Fill" for some reason lately ("magical", "incredible", "amazing" and others are fun to read).

I don't think I would be too far off in assuming that Adobes implementation is likely based off the wonderful work of Dr. Paul Harrison. Dr. Harrison's PhD thesis just happened to yield the code that gave GIMP users the Resynthesizer plugin, which has provided us a "Content Aware Fill" for quite a while now (since before 2005).

If you've ever spent any time with the Clone Tool or Heal Tool in GIMP, and haven't tried out Heal Selection with Resynthesizer, then you might find the results of this filter very helpful.

The rest of my GIMP tutorials can be found here:
Getting Around in GIMP

What does it do?

The basic premise behind Resynthesizer is that it will sample image textures from a given area, and will build a new texture from that data. The texture will be combined from segments of the selection in a random orientation. The plugin will also clean the feathering between segments to maintain a relatively consistent contrast, thus minimizing its visual impact on the result. (This is a very simplistic explanation of what's actually going on under the hood - there are some neat pixel neighborhood comparisons going on, actually).

So to illustrate, here is a small 100 × 100 px sample of some water:

A small water sample

Now running Resynthesizer on a larger canvas area (450 × 450) will yield:

After Resynthesizing

This produces a pretty nice new texture! Random enough that there are no discernible repeating patterns that would give it away as a rebuilt texture.

An interesting use of Resynthesizer for seamless textures explored by GimpChat.com user lylejk is described in this thread, for example.

Now, creating good seamless textures is handy, but what does it have to do with healing selections in our image? Well - this same code can be used to fill-in areas of our image from data nearby. This is exactly what the companion Script-Fu "Heal Selection" will do, with some neat options...

Heal Selection

So let's have a look at healing scenes from an image. Once you've installed the plugin and associated Script-Fu, the option will be available from:

Before we fire up the command, though, we just need to look at a couple of things. I'm going to use a cropped portion of a larger image to demonstrate:

Let's remove this person from the beach

Now, the normal way you might use Heal Selection to remove this person would be to create a selection around them, and then just run the command. So let's first create a selection around the object we want to remove.

I've colored in the selection to make it more visible

With our selection made, we could then start the Heal Selection dialog.

Before we run this, command, though, let's have a quick look at the options and what they mean.

The Context sampling width will allow you to choose how far out from the edges of your selection to sample for textures to fill with. In my example I am saying to use the surrounding image up to 75 pixels away from the edges to create the new texture that will be used to fill up my selection.

Selection to Fill and the 75px surrounding area to get the texture from.

If I run the command now, then what will happen is that the script will use the surrounding 75px area to build a texture to fill in my selection, and it will randomly fill the selection area for me.

So let's see how good of a job it will do with these (mostly) default settings:

Holy crap! (mouseover for original)

Well, wow. This has certainly done an admirable job in my opinion! The rebuilt portion of my selection contains enough randomness to blend very nicely without any artifacts to give it away.

The thing you really want to take away from this section is the use of the Context sampling width to adjust how far away from your selection to sample for the new texture. If you make the value too small for a large selection area, it may end up looking funny (due to having a small amount of pixels to sample from in order to fill a large area).

Well, we certainly can't stop here - so let's experiment even further with the entire image:

Beach Erosion by Kevin Dooley on Flickr (CC-BY)

I have already removed the person in the lower right frame of the image, and I will also remove the people in the water, and the man on the shore. I'll leave the small girl there for aesthetic reasons.

You can also select multiple areas to have Heal Selection fix them all at once if you'd like.

For instance, I will choose the man in the red shirt and the people in the water at the same time (don't forget to remove their shadows and reflections in the water):

I used a 50px sampling width, and selected both the people in the water and the man in the red shirt at the same time (mouseover for original)

If you want to be able to quickly visualize the sampling width, once you've made your selection you can grow it to the width you want using:

then just Undo the Grow to get back to your original selection.

Heal Selection works wonders provided that you have a relatively consistent type of surrounding textures to work with. It even does wonders where you might not think it would at first.

While playing around for this tutorial, I said "What if I wanted to remove the house in the upper right of the image?"

Well, turns out Heal Selection did a dandy job of fixing that right up for me:

Wow. No more house! (mouseover for original)

One other quick example, a nice serene beach scene (try guessing where the changes were made before you hover your mouse over it):

A nice quiet(?) beach scene... Beach Umbrellas by Randy Pertiet on Flickr

A Few Tips

There's a few things to keep in mind when using Heal Selection that might help you to achieve better results.

First is that depending on the geometry that you want to heal, the Sample from options will dictate from where around your selection you want to source pixels. The default will be All around, but the other two options can come in handy: Sides and Above and below.

This will restrict Heal Selection to choosing pixels only from either the sides, or above/below the current selection. I've found this helpful especially in places where there is a hard delineation in an image (trying to remove an object that crosses over the horizon, for instance).

I've used these options oftentimes in order to only partially remove an object from a scene (as part of a longer workflow). This is very handy around areas with a hard delineation.

For instance, remove the lower half of this telescope post to rebuild the brick wall behind it:

As part of a workflow, I only want to remove the post in front of the brick at the moment

Here is using the default Sample from: set to All around:

Notice the inclusion of part of the post in the lower shadow area of the selection. (mouseover for original)

Now if I restrict the sampling area to only the sides of the post, I will only get rock textures to create my fill textures with, which is much more desirable here:

Restricting the sampling to only the sides of the selection (mouseover for original)

This yields much better initial results for me to work with. The same method can be used for vertical edges as well (in which case you'd only sample from above and below).

A second tip that might be helpful is that you can run Heal Selection multiple times if needed in an area. Because the textures are being sampled from the image nearby, the overall grain/quality of the results will be close enough to the original to not stand out (though this entire process is very destructive to an image - so keep that in mind).

For instance, if I wanted to touch up the results from removing the post just a little more, I might want to remove the shadowed rock along it's top edge. No problem - just re-select a new area, and run Heal Selection one more time:

Second healing run to remove the shadow on the stone (mouseover for original)

Healing Panoramas

Of course, just like Adobes "Content Aware Fill", we can also use Heal Selection to extend areas of panoramic images where we are lacking source material. For instance, this is not an uncommon result of some panoramas:

ku skyline panorama by David DeHetre on Flickr

Using Heal Selection, we can have Resynthesizer rebuild the missing data for the sky and grass, using the nearby image texture data. In this case I just selected the black background areas, then grew the selection a bit to overlap the actual image.

Then, because I wanted to just sample from areas directly above the grass, and directly below the sky, I set Sample from to Above and below.

Heal Selection with sampling from above and below

This can be especially helpful to those creating 360° panoramas, and either don't capture a full, clean nadir or need to remove a tripod leg... (I had previously written about this image in my post On Little Planets).

I had captured the full nadir, but needed to remove tripod legs...

With the nadir frames missing, I can just run Heal Selection on the center of this panorama (2048 squared, set the sample width to 75px), resulting in:

Rebuilding grass for this nadir is a strong point of Resynthesizer

Rebuilding a random grass texture to fill in this nadir is an especially strong feature of Heal Selection, and it shows here.

Another note is that Heal Selection will not fill an area with texture if it is transparent. So if you have a layer with transparency in areas that you want to fill, you'll have to remember to remove the transparency from that layer first:

In Summary

Heal Selection/Resynthesizer is a killer filter. Due to the nature of GIMP plugins and scripts, it might not be immediately obvious to a new user that this is even an option. (We really need a better method of managing scripts and plug-ins in GIMP).

So here is a small list of links that might help:

The Resynthesizer page by Paul Harrison
Resynthesizer entry at the GIMP registry
(There is an updated version now. Find it here!)
Windows installers for various GIMP plugins (including Resynthesier/Heal Selection)
Partha Bagchi also maintains a few different GIMP versions and plugins
(I use his builds personally - particularly my native OSX version)

[Update 2012-12]
Over on the GimpChat forums, I walked through how I would approach an example by "slicing" up the area to heal based on surrounding textures. I thought it might be helpful to others who read this post, so check it out here!

Where it Works Well

Anywhere you have a need to fill from a rebuilt texture, and have at least something nearby to sample from. This could be mitigated through creative use of layers to bring appropriate textures closer to your healing area.

Large areas of similar texture, where you want to remove objects. Sky, grass, walls, etc. work very very well to resynthesize textures to heal your selections.

Where it Needs Help

Edges can be problematic to get an automated healing around. This can be mitigated by creative selections and directions for sampling, though. If your only other option is a lot of time with the clone and heal tool, then this method shouldn't be any harder or take any longer. (ie: try this first to help you get on your way to a clean heal).

If you've had a chance to try this filter out, please feel free to share the links with us in the comments below!

Embedding Fonts with CSS and Base64

2012-08-17T13:54:00.000-05:00

Relevant XKCD

One of my other hobbies besides photography happens to be web programming. I've been doing it in one form or another for many, many years (anyone else remember the first time animated .gifs were cool?). As a hobby it has been a ton of fun, and many of the newer capabilities just make it more so.

Anyway, in case you hadn't noticed I at least took the time to use interesting and pretty font faces on this site (in my opinion). It was literally hours of agonizing over different choices, weights, faces, readability, etc. I finally settled on two main fonts for this site that I thought worked reasonably well together:
Yanone Kaffeesatz for titles, and
Philosopher for my body text.
Google web fonts is a lifesaver here!

So I'm going to skip over a history of web-safe fonts and my using them on this site. Instead, I'll talk about the new Getting Around in GIMP page.

In a nutshell, I wanted to re-design the page to be cleaner and easier for people to use (and prettier as well). One of the things I wanted to try was a neat text-shadow css effect on the headings for each section. The problem was that I wanted to use a free font that I thought would look neat: Bazar, Medium by Olinda Martins.

The problem is that nobody was hosting it for direct embedding. I tried uploading it to Google Drive and directly hotlinking to it from here, but Firefox wanted nothing to do with it. I host this blog on Blogger, and there was no way for me to upload a .ttf font file here, either.

Then it struck me like a bolt of lightning...

I have seen a url for an image be given as a data: reference with base64 encoded data being attached. I figured if it worked for image data, maybe it would work for fonts as well! (I did something similar to this with my Chrome extension Patr for downloading your list of all your photos on Flickr).

If you're not familiar, this is what a base64 encoded data URI will look like:

<img src="data:image/png;base64,<BASE64 ENCODED DATA>" />

Now, could I do the same thing when I define the URL for the font I want to embed? Yes. Yes I can!

To embed a font directly into your CSS:

@font-face{

 font-family: Bazar;

 src: url(data:font/ttf;base64,AAEA… ) format('truetype');

}

Call me a geek, but that is seriously cool. No longer do I need to worry about hosting a ttf, otf, woff, eot file somewhere and embedding it - I can embed it directly into the CSS for my page!

To illustrate what I mean, I have just inserted the Bazar font into an inline style sheet in this post. If your browser supports it, you should be seeing it rendered here:

BAZAR, MEDIUM EMBEDDED!

And if you're curious, here is the effect I wanted to use with text-shadow:

BAZAR, MEDIUM EMBEDDED!

And the obligatory embossed effect:

BAZAR, MEDIUM EMBEDDED!

This works fine with Firefox/Webkit, and fails quietly on IE8 (don't have any other versions to test). So - if you're in need of embedding a font file directly on your page, and don't have a host for it, this is certainly an option!

And, sorry for the seriously off-topic post. I just figured I would put this here for anyone else searching for a similar solution (and for myself later when I am sure I will forget I even did this).

More Averaging Photos (Martin Schoeller)

2012-08-08T15:48:00.001-05:00

I can't get enough of playing with this image averaging, it seems.

Reddit user Janne mentioned in a post of mine about the idea of "Averageness", and linked to the Wikipedia page describing it. This got me thinking about trying these commands on something like faces.

So I needed a set of faces that were all semi-similar enough to create good averages with. Well, if you haven't seen the work of photographer Martin Schoeller you are missing out! He has a series of close-ups that are shot with very similar lighting styles and compositions of famous people (and not-famous), that is simply mesmerizing to see.

So I grabbed some random images to try this out with:

My random assortment of Schoeller close-ups.

In GIMP to prepare the images all I did was aligned the images based on the subjects eyes (tried to get them on the same level, and the same distance from the centers).

At that point the fun begins! All I had to do was pick and choose whose faces I thought would be fun to see averaged together. To begin with, here is the image matrix from above, but with averages calculated for each row and column appended to it:

Blends averaged on rows and columns.

What I learned from this matrix is that George Clooney and Justin Timberlake make attractive composites (and that Mickey Rourke makes them very scary).

So first I went ahead and just blended all of these faces together to see a total composite average image:

Composite of all of these portraits blended.

While that is interesting, the things that made me laugh out loud were the composites of faces I knew pretty well being put together in funny ways.

For instance, what about a simple two image blend with Harry Potter (Daniel Radcliffe) and Hermione Granger (Emma Watson)?

Harry Granger or Hermione Potter?

Another funny two person blend was George Clooney and Christopher Walken...

George Walken or Christoper Clooney?

Well, what about mixing it up even more? How about just the ladies?

Jolie, Watson, and Blanchett...

Or what about just the younger actors in the list?

Ledger, Radcliffe, Timberlake, and Watson

Or how about the more "distinguished" older gentlemen?

Clooney, Eastwood, Murray, Nicholson, Walken and Rourke.

And finally my personal favorite so far, all of the "dreamy" men as one composite:

My personal favorite - Clooney, Ledger, Timberlake and (Kelly) Slater

Honestly, after setting up the command and getting the images the same size and aligned, it was super easy just to drag and drop the ones I wanted to see mixed up. The variations were way too funny, and I was genuinely surprised at how some of them turned out.

Anyone game to try blending other series and sharing them?

ImageMagick Average Blending Files

2012-08-03T17:27:00.001-05:00

Back in 2003 Xeni Jardin posting on boingboing highlighted artist Jason Salavon's work involving averaging together Playboy centerfolds by decade. The works he produced were haunting and beautiful to me (and a little geeky, too).

Jason Salavon - Every Playboy Centerfold, The Decades (normalized) 2002

Those four images represented the four decades from the 1960's to the 1990's (left to right). The prints were made with code that he wrote in C on Unix based SGI machines.

These "Amalgamations" (to coin his phrase), are simply stunning to me. So in the sincerest form of flattery, I had to see if I could replicate the results in some fashion with more modern tools (Jason began his work blending the Playboy images back in 1997).

My first thought was to approach this using GIMP. After a few tries and experimenting I found that a better approach would be to use ImageMagick instead. It's just better suited for this type of processing.

Let's have a look at the results before we delve into the ImageMagick commands...

My first attempt was to replicate the results against all of the 1960's centerfolds to see if I could attain something similar (to validate the process):

Results after averaging all the centerfolds.

After running Levels → Auto in GIMP.

Overall I was pretty happy with the results!

In my case I rotated any images that were wider than they were tall to maintain the aspect ratio. While I was testing out the commands I also thought it would be neat to allow the horizontal images to remain that way, and to pad the dimensions to create a square result:

Not rotating the horizontal images, and centering/squaring produces this.

The result here is interesting because we also get a feel for the frequency of using horizontal spreads vs. vertical.

So I can confirm that the method appears to work. Just how do we do this?

ImageMagick Commands

There are two major things to consider when approaching this process:

Are the image sizes the same? If not we will have to resize and expand
Do you want to rotate horizontal images, or leave them as is?

These two options will have an influence on how we approach the commands we use. I will assume (to be safe) that none of my images are quite the same size, and that I do want to rotate wider images.

I also scanned through my images and determined that a the smallest image I wanted (or had available) would be 792x1728 - I will want to shrink all images that are larger than this to fit.

Here is the full command I used:

convert -background transparent *.jpg -rotate "90>" -resize "792x1728>" -gravity center -extent 792x1728 -evaluate-sequence mean output-rotated.jpg

ImageMagick commands are lately best read portion by portion as they operate on your images, so let's break this down:

convert -background transparent *.jpg

Start the "convert" command up, and set the background color to transparent for my operations. Also, load up all of my jpg files into memory.

-rotate "90>"

If any of the images are wider than they are high, rotate them 90°.

-resize "792x1728>"

Any images larger than 792x1728 should be resized down to fit in these dimensions (maintaining aspect ratio).

-gravity center -extent 792x1728

Keeping the images centered, extend the images so that they fill up the 792x1728 space (the padding will be filled with the background color - transparent).

-evaluate-sequence mean output-rotated.jpg

Finally, once all the images are now rotated and the same size, do an arithmetic mean on their pixels (average), and write the final output to output-rotated.jpg

Another version of the command above can be modified if I do not want to rotate any horizontal images (to create the square version above):

convert -background transparent *.jpg \
-gravity center -extent 1728x1728 \
-evaluate-sequence mean \
output-non-rotated.jpg

In this case I am not rotating anything, and padding all my images with transparency up to the extent size (1728x1728), then running the average. It should be noted that 1728x1728 is as large or larger than all of my images (otherwise a resize would be required).

More Results

With the command line stuff out of the way, let's take a look at some more interesting results I obtained playing with these options!

Continuing with the theme of Playboy centerfolds, here are the 1970's:

1970's normalized.

And what would decades of Playboy covers look like? (Honestly, the hardest part of creating these at this point is finding a decent archive of Playboy covers to use as source material... I would almost consider springing for the Playboy Cover to Cover Hard Drive - Every Issue From 1953 to 2010 if it wasn't so damn expensive!)

1950's

1960's

1970's

1980's

1990's

It would appear that over the years Playboy finally dialed in their title placement, as well as their model placements on the cover (they really got there in the 80's best, I think).

Enough with the Playboy, I think. At this point my keywords are going to make my blog look like a spam site. So let's switch over to something slightly less salacious, and have a look at all the covers of the Sports Illustrated Swimsuit issue! :)

Every cover of the SI Swimsuit edition through early 2000.

And let's finish up with something just a little more classy. Alberto Vargas may be the most famous name in classic Pin-up paintings, and interestingly the combination of all of his paintings is a little more tame. I came across a nice archive of his work at The Pin-up Files website.

All of Alberto Vargas' paintings at once.

Gil Elvgren is another famous name in Pin-up artwork. Here's a taste of what he produces when combined:

Gil Elvgren's works combined

And for something just a little more contemporary, here is a blend of all the headshots from Peter Hurley's sample portfolio of Leading Ladies:

Peter Hurley's Leading Ladies

In each of these I brought the final output images into GIMP for adjusting levels. I normally just went ahead and let the Levels tool auto-adjust for me, but I did look at stretching contrast (+hsv) to see the results. In the end I think I preferred the Levels → Auto the best.

Give it a try and let me know how it turns out, I'd love to see some other cool visualizations of sets of photos in this way!

Also, if Jason Salavon ever reads this - thank you for turning me on to a great and interesting processing technique! Your results are amazing!

I've also tried my hand at blending a bunch of celebrity faces together as well, have a look here:
More Averaging Photos (Martin Schoeller)

If You Can't Fix It, Fake It

2012-08-01T15:45:00.000-05:00

An interesting side effect of posting the Getting Around in GIMP tutorials has been meeting new people who are looking for help with images in some way.

I think my most heartwarming interaction was a mother who wanted some images fixed of her newborn son that had discoloration in the skin. She wanted the skin tones to be matched so that the discoloration wouldn't be so prominent/noticeable. I was happy to take a few minutes to try and help out (the fix was relatively straightforward), and it was extra meaningful to find out that her little son did not live long after the photos were taken. (That last part in particular was especially poignant to me, being a dad myself).

A recent interesting problem was posted to me where someone had created artwork to be turned into a custom wood puzzle as a gift.

The problem that needed to be fixed was that a photograph created by the puzzle maker was taken with a small P&S camera with (likely) auto white balance under fluorescent lighting.

White balance is absolutely wonky here in this original photograph.

The images came out with a really nasty orange tint to them. So the question posed to me was if my recent post on color curves could help fix them to match the original artwork.

In this case, no. Color correction with this heavy of a cast was not going to really be possible. The best effort from the person who sent them to me was to desaturate heavily to try and get the colors of the train closer to the original.

Other attempts at color corrections were too violent for the jpeg to handle, and nasty posterization and banding were very evident. So curves/adjustments were out of the question.

After attempting a desaturate to bring colors more in line with the original artwork.

The desaturation helped with some of the colors, but as a global operation it severely dampened the vibrancy of many of the other colors.

Below is the original artwork prepared by the person who contacted me. This original artwork represents the true colors that they wanted to match in the photograph above.

The original artwork that was printed and mounted onto the wood.

The thing that finally clicked for me in this case was when I was looking at the high quality original artwork as a reference for colors. I was trying to think of a good way to match the colors when I caught myself saying "If only I could just map this original artwork into the photograph".

That's when a different part of my brain said: "Well, you can map this original to the photograph: use Hugin!".

I loaded up both the original artwork and the photograph into Hugin. I set the photograph to be anchored for position, and created control points between the artwork and photo.

Once I had all my control points, I optimized the artwork (making sure that none of the checkboxes for the photograph were checked - so only the artwork would be distorted to fit).

I did a set of standard optimizations before deciding that the most important ones are for x-shift and y-shift (allowing skewing of the image).

I was pretty happy with the results, and am amazed yet again at the neat things you can do with some incredible free software:

After remapping the original artwork to fit into the photograph (mouseover for original)

I was glad to be able to help someone out, even if the path to getting there was a little unconventional (all that mattered was the results in this case). In the process, I learned a couple of neat new tricks to keep in my bag for the future!

How Not to Make a Wee Planet in GIMP

2012-07-22T21:58:00.000-05:00

Wee Mobile Alabama (Cathedral of the Immaculate Conception) by Pat David

I was fooling around again recently with some "Wee Planet" stereographic panoramas (previously mentioned here), and came across some tutorials on doing this with Photoshop/GIMP using a Polar Coordinate distortion filter.

While this will give you a similar result to what you might see from a true stereographic projection, it is not quite as aesthetically pleasing I feel. This is because the Polar Coordinate transform is not a true stereographic mapping, and will tend to squish objects in a very non-pleasing way.

I've assembled a few examples to demonstrate the effect and for comparison. The examples here are from Flickr user gadl (Alexandre Duret-Lutz) and are Creative Commons BY-NC-SA. Thank you Alexandre for making these available!

And really, what better examples to use than from "La Ville-Lumière", Paris!

Sacré coeur

I once had a wonderful lunch with some friends not far from here in Montmarte...

Starting with the equirectangular panorama projection below:

Equirectangular Sacré coeur by Alexandre Duret-Lutz

Running the image through

Filters → Distorts → Polar Coordinates...

in GIMP yields the results below...

Polar Coordinate Distortion in GIMP

Notice how the Basilica is squashed in this view. In fact, all of the structures above the horizon have been seriously shortened. Because the polar coordinate transformation is essentially just remapping the image to a circle, notice also that the distance from the center to the outer edges is the same as the height of the source image.

Loading the image into Hugin, though, allows us to remap the output to a true stereographic projection:

Stereographic mapping in Hugin

The results here are much more aesthetically pleasing, in my opinion. The local angles are preserved, and the relative prominence of objects above the horizon become much larger.

Here's another view:

Montmartre s'éveille by Alexandre Duret-Lutz

Notice how squished the lamp post and Basilica are...

This looks much more like a tiny planet.

Here's a view of Notre Dame at night...

Notre Dame de Paris by Alexandre Duret-Lutz

Polar Coordinate Distortion in GIMP

Stereographic mapping in Hugin

Separation light by Alexandre Duret-Lutz

Polar Coordinate Distortion in GIMP

Stereographic mapping in Hugin

In each of these cases I personally much prefer the stereographic projection.

The only drawback of this method is that it requires a full 360° × 360° full panorama of your scene (for best results). This is not always practical if you're shooting on the fly. If the scene you want to capture is intended to be a little planet, it's definitely worth the time to capture a full panorama to map stereographic.