[WIP] Add 3D support to sobel

[dcz-mpipz]

Added generalized version of Sobel edge detection in a particular direction. Accepts N-dimensional images, using h=[1, 2, 1] and h'=[1, 0, -1]. Only tested for 3D images. however.

Description

This change provides a way to implement a full Sobel operator, as noted in #2247

The sobel implementation is directional, and provides a function generating kernels. The values of sobel function can be greater than 1, despite normalizing the matrix. Dimensions above 3 were not checked, since I don't know what to expect and how to write tests.

Adding a full Sobel shouldn't be difficult, e.g. by analogy to 2-dim Sobel:

        res = numpy.empty_like(image)
        for axis in range(image.ndim):
            res += sobel(image, axis, mask) ** 2
        res /= np.sqrt(2)

I did not implement that, since I don't know whether sqrt(2) is a suitable factor for higher dimensions.

Checklist

• Full sobel
• Unit tests for fimensions above 3

For reviewers

(Don't remove the checklist below.)

• Check that the PR title is short, concise, and will make sense 1 year
  later.
• Check that new functions are imported in corresponding __init__.py.
• Check that new features, API changes, and deprecations are mentioned in
  doc/release/release_dev.rst.

[pep8speaks]

Hello @dcz-mpipz! Thanks for updating the PR.

Cheers ! There are no PEP8 issues in this Pull Request. 🍻

Comment last updated on September 01, 2017 at 12:56 Hours UTC

[emmanuelle]

I'm not sure this is the right normalization but on the other hand the normalization of the 2D kernel seems weird (it does not correspond to the definition of the Wikipedia web page for Sobel)

[dcz-mpipz]

This was just a guess, based on what other kernels in edges.py look like. It's more of a placeholder.

[emmanuelle]

From git blame I think it was @ahojnnes who wrote the kernels with the normalization.

[emmanuelle]

I think you would need to modify also the _mask_filter_result function so that it uses a structuring element of the same dimension as result or mask.

[dcz-mpipz]

You're right, I'll try to get around to it soon. Should that be a separate pull request?

[emmanuelle]

No, it should be done in the same branch, and as you push new commits they will automatically appear in this pull request.