skimage.measure.find_contours returns results outside masked areas

[PierreRaybaut]

Description:

The skimage.measure.find_contours function sometimes returns contours which are entirely outside the input array masked areas, and thus that should be filtered out. This was reported in the context of DataLab (see the associated issue).

In DataLab, the "detection contour" feature is based on this function and, following a bug report, a workaround was added to simply filter out all contours which coordinates correspond to masked pixels:

    contours = measure.find_contours(data, level=get_absolute_level(data, level))
    coords = []
    for contour in contours:
        # `contour` is a (N, 2) array (rows, cols): we need to check if all those
        # coordinates are masked: if so, we skip this contour
        if isinstance(data, ma.MaskedArray) and np.all(
            data.mask[contour[:, 0].astype(int), contour[:, 1].astype(int)]
        ):
            continue

Way to reproduce:

The test image may be found here.

The test script which reproduced the issue (before it was fixed by adding the functions mentioned above) is here.

Please also take a look at the associated issue which describes the original problem.

Version information:

3.11.5 (tags/v3.11.5:cce6ba9, Aug 24 2023, 14:38:34) [MSC v.1936 64 bit (AMD64)]
Windows-10-10.0.22631-SP0
scikit-image version: 0.22.0
numpy version: 1.26.1

[soupault]

Hello @PierreRaybaut and thank you for reporting!
Generally, scikit-image does not support masked arrays, so I would not consider this as a bug. In the context of this issue, however, it would be appropriate to clarify our stance in the documentation, e.g. https://scikit-image.org/docs/dev/user_guide/numpy_images.html .