Ensuring repeatable global histogram results

[nullpainter]

I've noticed that output images are very subtly different after histogram equalisation is applied with the default global options.

As a pretty basic test, the following code:

for (var i = 0; i < 5; i++)
{
    using var image = await Image.LoadAsync(@"c:\temp\input.jpg");
    image.Mutate(c => c.HistogramEqualization());
    await image.SaveAsync($@"c:\temp\output_{i}.png");
}

produces files all with slightly different sizes:

06/11/2020  05:48 pm         3,445,866 0.png
06/11/2020  05:48 pm         3,429,505 1.png
06/11/2020  05:48 pm         3,449,050 2.png
06/11/2020  05:48 pm         3,434,962 3.png
06/11/2020  05:48 pm         3,434,531 4.png

I haven't looked into the code too much, but I'm assuming there's some sort of random sampling going on. Is there a way to ensure repeated applications of HistogramEqualization produce consistent results?

The background is that I'm applying a CLUT to an intensity range of a greyscale image, after histogram equalisation has been applied. The subtly different results between runs is resulting in noticeable differences in the CLUT-enhanced output and it would be good to avoid this if possible.

[antonfirsov]

@nullpainter this is weird, I can't see any randomness in the processor implementations, the outputs should be deterministic.

Can you double check this with an image diff tool, and if you can prove there is a pixel difference:
open an issue with template filled + images attached (or even better a repro code doing a pixel comparison)?

[JimBobSquarePants]

I managed to recreate this so @nullpainter please raise an issue.

[nullpainter]

Will do. Would you like me to make a little repro project?

[nullpainter]

I'll make one anyway, just for fun.

[nullpainter]

Done (#1416). Thanks!

[brianpopow]

The issue is now fixed with the release of v1.0.2