This is a tool for simulating and/or correcting for the main three types of colour blindness, protanopia, deuteranopia, and tritanopia. The first two of these, known collectively as red-green colour blindness, are the most common, and are X-chromosome linked, meaning they predominantly affect males. Tritanopia, or blue-yellow colour blindness, is much is rarer, but age-related lens yellowing can be seen as a mild form of it.
The tool can either process one-off source images, or be used to generate lookup tables, for use in real-time applications, e.g., as a post-process in a shader. Alternately the functions in CBLut.* can be incorporated directly into your own code.
The default form of colour-blindness correction supplied operates in LMS space, and primarily shifts hues in affected areas. For instance, common colour-blindness advice for protanopia and deuteranopia is to avoid the classic red/green colour combination, and prefer magenta/green, and this approach effectively transforms the former into the latter.
Also implemented is the commonly-used approach of Fidaner et al., which redistributes error in the affected channel to other channels in RGB space. This is typically referred to as "Daltonisation", although that is really a more general term for any form of colour correction. It has the effect of brightening and desaturating the image in affected areas.
In my tests the hue-based correction does a better job than Fidaner, however, I'd suggest trying both to see which best suits your particular scene type, as results are dependent on the distribution of source colours.
The LUTs are in 32x32x32 RGB cube format, represented as 32x1024 2D images, as this is generally a good compromise between fidelity and size. (This can be changed by modifying kLUTBits in the source.) If you're only interested in the LUTs, pregenerated versions can be found in the luts directory.
Identity
Simulate: P/D/T
Correct: P/D/T
Daltonise: P/D/T
The tool can also be used to apply arbitrary LUTs to source images. For instance, by generating an identity LUT, then applying arbitrary image processing operations to that LUT (say in Photoshop), you'll get a LUT that can be used to apply the same operations to any image with a single texture lookup.
The strength of colour blindness can be specified via "-m float", to reflect the common case of anomalous vision, namely, one cone type having reduced sensitivity, rather than being completely missing. In the case of the custom correction, a mixed strategy that initially brightens the affected channel and then switches to hue shifts is used. This is a balance between strengthening the affected channel, which works well for small losses, and the hue shift, which works best when the loss is large enough that compensating directly would lead to out-of-band luminance values.
There are numerous test images for red-green colour blindness, most famously the Ishigara coloured-dot diagrams. A selection of these are provided in the tests directory. Blue-yellow colour-blind test images are on the other hand much rarer, so cblutgen implements a strategy for converting the former into the latter -- see the -r option. Note that this is not as simple as just swapping channels, as doing that affects the image white point.
Finally the tool supplies two colour blind-savvy false colour (luminance -> rgb) maps, via the -c option. It also supports three more standard maps, and applying any 256-wide png as a map. See cblutgen -h for more info, and ColourMaps.h for references.
Viridis (decent)
Cividis (optimised further, a bit plainer)
To build and run the tool, use
c++ --std=c++11 CBLuts.cpp ColourMaps.cpp CBLutGen.cpp -o cblutgen
Or, include these files in your favourite IDE, build, and run.
To generate simulated and corrected versions of the supplied test images, along with markdown-style results files, run the supplied "generate" script. (Currently unix-style OSes only.) The results can be found in the 'out' directory.