[Insight-users] Object-based processing

[Jordi Inglada]

Dear all,

I would like to do some processing (say, computing statistics) on an
image, but not in the classical pixel-wise or neighborhood-wise way, but
rather on segments (or objects). For example, I would like to compute
the variance inside every region obtained after a watershed
segmentation.

I understand that I could use masks (spatial objects or image masks),
but I guess this may not be very efficient, since I do not want to
compute this for a single region of the image, but for all regions of an
image partition.

Of course, I would like to have streaming and threading in my
pipeline ;).

I have been looking at the library API and I do not see any
straightforward approach to implement this kind of processing.

I have been using the "NaryFunctorImageFilters" and I have even created
a class for NaryFunctorImageWithNeighborhoodFilters (when the processing
is not only pixel-based, but with a sliding window). Colleagues of mine
have implemented a "WithOffset" version in order to compute texture-like
features.

I guess it would be nice to have something like a neighborhood iterator
but for connected components in the image (segments or objects). But
before diving into such kind of implementation, I would like to have
advices from experienced ITK users about this problem.

Thanks in advance for your help and feedback.

Jordi 

-- 
CNES - DCT/SI/AP - BPI 1219
18, avenue Edouard Belin
31401 Toulouse Cedex 09 - France
Tel. +33.(0)5.61.27.33.97 - Fax. +33.(0)5.61.28.31.09