ENH Add crop pipeline

[caspervdw]

This adds a cropping pipeline that correctly changes the frame_shape attribute. I decided to keep the API the same as that of skimage.util.crop. There are many other pipelines that we could implement (for instance, rewrite normalize and to_rgb from pims.display, but I think it is a good idea to stay close to skimage and where possible, just import the functions instead of redoing it.

A general approach for that would be testing the function on the first frame at __init__, recording the changed attributes, and exposing them automatically. That might save us a fair amount of work, and the performance loss would be not too bad. This would be a pims functionality, as it would be specific to images (frame_shape, pixel_type, etc.).

In this case of crop, I did end up copying over the code.

As it makes use of soft-matter/slicerator#21 , we need a slicerator release to get the tests working.

[danielballan]

I'll put a slicerator release on my todo list for next week unless you can get to it sooner.

[caspervdw]

@danielballan I'll do the release now

[caspervdw]

I would like to start adding some more pipelines, but this has to go in first
Do you agree with exposing pipelines with the same API as skimage, importing where possible?

[tacaswell]

What is the current state of this?

[rbnvrw]

Hi @caspervdw, we'd like to add this very old PR to the new pims release! Do you have time to update the PR or should we do it?

[caspervdw]

@rbnvrw I merged master and attempted to fix an incompatibility with numpy

[caspervdw]

@rbnvrw It seems that slicerator has become incompatible with the newest numpy.

[nkeim]

Thanks @caspervdw ! I'm not sure the problem is the newest numpy—our "legacy" Travis environment uses the ancient numpy 1.8.2 and still had similar errors as the other environments, such as this:

58ERROR: test_on_frame (pims.tests.test_process.TestCrop)
559----------------------------------------------------------------------
560Traceback (most recent call last):
561  File "/home/travis/build/soft-matter/pims/pims/tests/test_process.py", line 44, in test_on_frame
562    assert_equal(self.pipeline(self.rdr[0]), self.first_frame)
563  File "/home/travis/build/soft-matter/pims/pims/tests/test_process.py", line 57, in <lambda>
564    self.pipeline = lambda x: crop(x, ((5, 32-26), (7, 33-27)))
565  File "/home/travis/mc/envs/testenv/lib/python2.7/site-packages/slicerator/__init__.py", line 641, in process
566    return cls(*(tuple([a] for a in ancestors) + args), **kwargs)[0]
567  File "/home/travis/mc/envs/testenv/lib/python2.7/site-packages/slicerator/__init__.py", line 480, in __getitem__
568    return self._get(indices)
569  File "/home/travis/build/soft-matter/pims/pims/process.py", line 93, in _get
570    return np.array(ar[self._crop_slices], order=self._crop_order,
571TypeError: list indices must be integers, not tuple

I may be able to take a closer look myself. Worst case, we can just hold this PR until a minor release.

[nkeim]

@caspervdw I think your commits fixed everything, except for a small mistake. The tests pass now. Could you please check that I got this right?

[caspervdw]

Thanks for spotting that! Everything is ok now.

[nkeim]

Great! Thanks @caspervdw for writing such a useful and logical feature! I'm glad it will get released.