Computer Science Colloquium

A contextual pattern represents an object that consists of multiple object primitives and contextual relations among those primitives. I will present the methodology and theory for automatic contextual pattern modeling for automatically learning a parametric model from multiple samples. The maximum-likelihood parameters of the model are learned via the Expectation-Maximization algorithm. The learned model characterizes both the appearance and the structure of the object, which is observed under various conditions (e.g., lightings, backgrounds, viewpoints, etc.). The theory has been applied to unsupervised image pattern extraction, improving pattern detection, video summarization and retrieval, texture modeling and synthesis, and automatic frequently asked question detection.

Pengyu Hong passed his Ph.D. thesis defense and will receive his doctorate from the Department of Computer Science at the University of Illinois at Urbana-Champaign in December 2001. He is now a post-doc in the Coordinated Science Laboratory at the University of Illinois at Urbana-Champaign. He is conducting research in the areas of multimedia information processing, machine learning, pattern recognition, human-computer interaction, computer vision and graphics, and bioinformatics.