EECS CSPL SEMINAR SERIES
WINTER TERM 1997

Yingnian Wu

Dept. of Statistics

University of Michigan

yingnian@stat.lsa.umich.edu

Tuesday, March 18, 1997
4:30-5:30 pm
1003 EECS

Texture Modeling by Combining Filters and Markov Random Fields

Abstract -
Statistical models for natural textures are important building blocks in the Bayesian paradigm for computer vision. This talk presents a class of Markov random field models for textures, with the feature that the local spatial interactions (or potential functions) are characterized by one-dimensional nonlinear functions of responses extracted by a set of filters. Methods for model fitting, stochastic simulation, and filter selection will be described. Some experiment results will be displayed. Some open questions will be discussed.

Based on the joint work with S.C. Zhu and D.B. Mumford.

Reference:

Zhu, S.C., Wu, Y.N., and Mumford, D.B., "Filter, Random Field, and Maximum Entropy (FRAME): towards a unified theory for texture modeling", to appear in IJCV.

Biosketch -
Yingnian Wu got his Ph.D. degree of Statistics in Harvard in Nov. 1996. He joined the Dept. of Statistics, Univ. of Michigan in Feb. 1997.