Improve support for ROIs, segmentation and labeling (2.0.0-beta4)

[leek]

Propose and refine a mechanism to represent segmentation and labeling in ImageJ.

[leek]

Segmentation proposal

[leek]

I've posted the whitepaper to the imagej dev mailing list and have received some useful feedback:

• Discussed use of existing shape cursors in imagelib to represent labels and agreed that the overhead they would entail is risky.
• Discussed adding cursors that iterate over the pixels in the perimeter of labeled objects - there are a number of measurements that only rely on the perimeter pixels. There are compact representations of labelings based on just the perimeter pixels (at least, for blobby objects).

I'm starting to code a reference implementation to prove out the concepts.

[leek]

I have implemented a preliminary version of labeling. I have also implemented the "All connected components" algorithm for labeling connected regions of a binary image and for labeling an intensity image using the watershed algorithm.

I'll wait for comments before resolving as fixed; I might work on the algorithms to bring them more into line with the others in imglib.algorithms.

[leek]

I will present my segmentation code during the Hackathon - slide deck and talk preparation.

[leek]

I have updated the labeling code in the imglib2-roi branch to work with the rest of imglib2 with the exception of the gradient watershed method. This is waiting on completion of FFT code in imglib2.