Sim Daltonism’s purple bug

Back in 2016, I converted the filter algorithm in Sim Daltonism 1.x to OpenGL shader code so it’d run on the GPU and have better performance. This also made it possible to bring it to iOS as a real-time filter for the camera. A bug slipped through however.

On a filtered color wheel we can observe that a region of deeply saturated blue becomes purple. This makes no visual sense: purple light is blue light + red light, and this purple area showed up even for protanopia (no red cones). This also wasn’t happening with version 1, and it wasn’t happening on the Color Laboratory on which Sim Daltonism’s filter algorithm is based.

I dug up the code for version 1 and tried to find differences. I did not have much success at finding accidental algorithmic differences. So I finally tried to clamp some values within the 0…1 range they were supposed to be in; theoretically the result of the calculation shouldn’t be out of this range but that doesn’t account for rounding errors in floating point calculations. GPUs are quite fast, but they sometimes sacrifice precision to achieve that speed. And that did the trick.

So now there’s a new version of Sim Daltonism with a corrected filter on both Mac and iOS. And I want to thank Lisa Charlotte Rost for signaling this bug after comparing with other color blindness simulation tools.

  • © 2003–2020 Michel Fortin.