ENH: support multiple labels in ContourExtractor2DImageFilter

[Leengit]

The previously existing function of ContourExtractor2DImageFilter is to trace the contour between regions of a 2-dimensional image defined by values that are higher and lower, respectively, than a specified contour value. To that we add a LabelContoursOn() option; when set then every distinct value in the 2-dimensional image defines its own region, and boundaries for all such regions are generated.

At present this is WIP. Please review for the API changes. The implementation works, but is not performant; it needs additional work.

PR Checklist

• No API changes were made (or the changes have been approved)
• No major design changes were made (or the changes have been approved)
• Added test (or behavior not changed)
• Updated API documentation (or API not changed)
• Added license to new files (if any)
• Added Python wrapping to new files (if any) as described in ITK Software Guide Section 9.5
• Added ITK examples for all new major features (if any)

[Leengit]

I force-pushed with performance improvements. It could be even better, but this is plenty good enough. I will change this Pull Request to non-draft once it has passed its GitHub checks.

[Leengit]

A next step for performance improvement could be to address the bounding box calculation for each label. Do we have something like skimage.measure.regionprops functionality, which computes bounding boxes for multiple labels all at once, in ITK? If yes, a future pull request could have this code make use of that. If not, a future pull request could add the regionprops functionality to ITK .

[Leengit]

This pull request adds a new ValuesAreLabels option to itkContourExtractor2DImageFilter. Is there an API document somewhere where that should be documented ... or are the Doxygen comment blocks within itkContourExtractor2DImageFilter.h sufficient?

[dzenanz]

Doxygen blocks are sufficient, as the documentation is generated from them.

[thewtex]

ValuesAreLabels -> LabelContours

[thewtex]

👍

[Leengit]

I just rebased to the current master branch and squashed the commits. This is ready for a final sanity check and merge.

[N-Dekker]

The implementation works, but is not performant

@Leengit Do you think the performance is good enough now? Just curious. 😃

I wonder if it would be possible/useful to make the iteration over the labels for (InputPixelType label : allLabels) the inner loop, instead of the outer loop. Do you think that could affect the performance in a good way?

Anyway I don't have concrete plans to use the filter right now (sorry), I'm just trying to help a little bit!

[Leengit]

Timings for the algorithm are now good, about 17x speed compared to the initial implementation, and beating the competitors such as HistomicsTK. The loop over labels as an inner loop is a good idea ... it would involve having separate lists of contours for each label being built up in parallel. We are going to go with the parallel constructions, but using a multi-threaded approach: issue #2343, which @dzenanz has assigned to himself, and is awaiting the merging of this PR.

@N-Dekker writes:

Anyway I don't have concrete plans to use the filter right now (sorry), I'm just trying to help a little bit!

Very much appreciated!