The main obstacle to using wavelet-based priors for segmentation (or any other priors for continuous-valued images, such as Gauss Markov models) is that they are aimed at representing real values, rather than the discrete (categorical) labels needed for segmentation. I'll show how this difficulty can be sidestepped by introducing real-valued hidden fields, to which the labels are probabilistically related. A (generalized) expectation-maximization algorithm can then be derived to perform "maximum a posteriori" segmentation under this model. Experiments on synthetic and real data testify for the adequacy of the approach.
This is an extended version of a recent oral presentation at the IEEE Computer Society Conference on Computer Vision and Pattern Recognition - CVPR'2005, San Diego, CA, June 2005.
Time and Place: Mon., Aug. 29, at 3:30 pm in 2534 Engr. Hall. *** NOTE SPECIAL DAY & ROOM ***
SYSTEMS SEMINAR WEB PAGE: http://homepages.cae.wisc.edu/~gubner/seminar/