Mahotas 1.1.0 Released!

Mahotas 1.1 Released

released mahotas 1.1.0 yesterday.

Use pip install mahotas --upgrade to upgrade.

Mahotas is my computer vision library for Python.

Summary of Changes

It adds the functions resize_to and resize_rgb_to, which can be used like:

import mahotas as mh
lena = mh.demos.load('lena')
big = mh.resize.resize_rgb_to(lena, [1024, 1024])

As well as remove_regions_where, which is useful for handling labeled images:

import mahotas as mh
nuclear = mh.demos.load('nuclear')
nuclear = mh.gaussian_filter(nuclear, 2)
labeled,_ = mh.label(nuclear > nuclear.mean())

# Ok, now remove small regions:

sizes = mh.labeled.labeled_size(labeled)

labeled = mh.labeled.remove_regions_where(
        labeled, sizes < 100)

Moments computation can now be done in a normalized mode, which is robust against scale changes:

import mahotas as mh
lena = mh.demos.load('lena', as_grey=1)
print mh.features.moments.moments(lena, 1, 2, normalize=1)
print mh.features.moments.moments(lena[::2], 1, 2, normalize=1)
print mh.features.moments.moments(lena[::2,::3], 1, 2, normalize=1)

prints 126.609789161 126.618233592 126.640228523

You can even spell the keyword argument “normalise”!

print mh.features.moments.moments(lena[::2,::3], 1, 2, normalise=1)

This release also contains some bugfixes to SLIC superpixels and to convolutions of very small images.

(If you like and use mahotas, please cite the software paper.)


Why Pixel Counting is not Adequate for Evaluating Segmentation

Let me illustrate what I was trying to say in a comment to João Carriço:

Consider the following three shapes:


If the top (red) image is your reference and green and blue are two candidate solutions, then pixel counting (which forms the basis of the Rand and Jaccard indices) will say that green is worse than blue. In fact, green differs by 558 pixels, while blue only by 511 pixels.

However, the green image is simply a fatter version of red (with a circa 2 pixel boundary). Since boundaries cannot be really drawn at pixel level anyway (it is a fuzzy border between background and foreground), it is not an important difference. The blue image, however, has an extra blob and so is qualitatively different.

The Hausdorff distance or my own normalized sum of distances, on the other hand, would say that green is very much like red, while blue is more different. Thus they capture the important differences better than pixel counting. I think this is why we found that these are better measures than Rand or Jaccard (or Dice) for evaluation of segmentation.

(Thanks João for prompting this example. I used this when I gave a talk or two about this paper, but it was lost in the paper because of page limits.)


NUCLEAR SEGMENTATION IN MICROSCOPE CELL IMAGES: A HAND-SEGMENTED DATASET AND COMPARISON OF ALGORITHMS by Luis Pedro Coelho, Aabid Shariff, and Robert F. Murphy in Biomedical Imaging: From Nano to Macro, 2009. ISBI ’09. IEEE International Symposium on, 2009. DOI: 10.1109/ISBI.2009.5193098 [Pubmed Central open access version]