Scaling and Artifact Problem in colorview

[johnnychen94]

Manually transfered from JuliaImages/Images.jl#882. The following contents are originally posted by @roflmaostc. This seems only an ImageMagick issue, using PNGFiles/ImageIO work arounds it.

---

Hey,

I observed a bug for which I didn't find an explanation so far. I wrote a Poisson noise function which adds noise to the image. If I inspect the final images with imshow, all seems to be fine.

However, only for the parameter value nphot=3, nphot=5 and nphot=15, if I save the image (or inspect them directly with colorview) these images are scaled and shifted to the left. On the right part, there are some artifacts

I cropped the images so that only intensities between 0 and 1 are possible.

Not working: nphot=3, nphot=5, nphot=15

One working example: nphot=10

Here is the full code example:

using Images, TestImages, PoissonRandom, ImageView

function my_poisson(img; nphot=100)
    nphot = float(nphot)
    max_intens = maximum(img)
    #scale image to maximum number of photons and apply poisson noise
    img_noisy = map(x -> pois_rand(x / max_intens * nphot), img)
    #rescale to initial intensity
    img_noisy = img_noisy ./ nphot .* max_intens

    #crop intensity
    img_noisy = map(x -> max(0.0, x), img_noisy)
    img_noisy = map(x -> min(1.0, x), img_noisy)

    return img_noisy
end


img = testimage("cameraman")

#i=3, i=5 and i=15 cause strange output
for i=1:15
    img_poiss = my_poisson(img, nphot=i)
    img_poiss_view = colorview(Gray, img_poiss)
    save(string("test", i, ".png"), img_poiss_view)
end

#this part works totally fine
img_poiss = my_poisson(img, nphot=15)
img_poiss_view = colorview(Gray, img_poiss)

imshow(img_poiss_view)

Edit: with different image size, the parameter, where the result is weird, changes.