Title: Classification of Regions using Belief Networks and Wavelet/Fractal Multiresolution Analysis

Talk by Dr.Marcus J. Sobel , Statistics Department - Temple University

Friday 12-06-02

Room: 322 Wachman Hall

Time: 1:30-2:30

Abstract: In most of the attempts to characterize data (images, signals, text, etc.) the prime concern is to extract descriptive features that provide significant information. This process is complicated by:

a) The necessity of combining or reweighting the large number of features under consideration
b) Accurately modeling the noise process implicit in their analysis, and
c) Taking into account the correlation between features.

We focus on characterizing spatial regions of interest in a supervised framework. We comment on how this might be extrapolated to classification in unsupervised frameworks. We employ two methodologies in our analysis:

a) Belief Network Models serve to accurately characterize dependence between features in region analysis.
b) Wavelet Multiresolution (Tree) Models serve to accurately characterize noise by modeling it on a 'level by level' basis. They also supply a natural framework for combining and reweighing features.

2:15 -2:30 Time for Discussion