Student Seminar Series - September 8, 2005
University of Minnesota
School of Statistics
College of Liberal Arts

Statistical Modeling of Images


Xiaoyan Li


Thursday, September 8, 2005
1:30 PM, 300 Ford Hall
Minneapolis, East Bank Campus

Refreshments at 1:00 PM
300 Ford Hall


Abstract

Image skeletonizing and vectorization are widely used techniques in object classification and shape detection. But the skeletonizing process consists of several stages, e.g. dichotomization, boundary pruning, propagation and further detailed reduction, each of which involves subjective parameter setting. Currently, there is no widely agreed upon skeletonizing algorithm that can provide definite single pixel skeleton result as needed in classification and other comparison. The situation is similar to the vectorization process where important parameters are selected by the programmer. These low level tasks are sequentially rigid and non reversible. i.e. mistakes made in earlier processing step will not have a chance to be corrected. To incorporate these uncertainties into the estimation of the higher level structure, we intend to model the skeletons as a function of hidden GMRF on all observed pixel intensities and use simulated annealing to get maximum likelihood estimate.