Make OffsetArray usable in edge.jl

[vis-florum]

Currently, when trying to detect edges with canny() for an image in an OffsetArray, error occurs because of an indexing issue in thin_edges_nonmaxsup_core!().
In specific, the problem seems to be from line 301 onwards because:
[...]
(height,width) = size(img)

for x = 1:width, y = 1:height
    (c = img[y,x]) == 0 && continue  # For thresholded images

[...]
the image is addressed by x and y which depend on the absolute size of the array.
Instead, one could maybe change it to use CartesianIndices of the image instead.

I would be great to be able to use OffsetArray and the CartesianIndices because both are extensively used throughout the Images.jl package.

Thanks for your great work!

[zygmuntszpak]

Thank you for reporting this issue. I am in the process of finalising a Canny edge detection routine here (JuliaImages/ImageEdgeDetection.jl#9) and can confirm that it works on OffsetArrays. I mainly just need to verify and document the API before I can register that package.

I have registered the package.

[knuesel]

@zygmuntszpak will this replace the implementation in Images.jl?

[zygmuntszpak]

@knuesel Yes, the intention is for the Images.jl package to use ImageEdgeDetection.jl and to depracate the current canny function etc. How to best manage this transition is still something we need to figure out.