The Promise and Perils of Near-Regular Patterns
Speaker: Yanxi Liu , Carnegie Mellon University
Date: March 8 2004
Time: 2:00PM to 3:00PM
Location: Open area in Vision Lab, 6th floor 400 Tech Sq
Host: Bill Freeman, CSAIL
Contact: Greg Shakhnarovich, gregory@ai.mit.edu
Relevant URL: ***** NOTE the unusual time (Monday) ****
ABSTRACT:
We are surrounded by near-regular patterns that are (1) man-made: textiles, architectures, walls, pottery, arts, website structures; (2) biological: cells, DNA structure, faces, brains, feathers, various types of gaits; and (3) natural: waves, leaves and sand patterns. With the increasing power of computer hardware, the lack of computer algorithms that understand the essence of near-regular patterns has become more and more obvious. This shortage is preventing autonomous computer systems from readily taking full advantage of regularity when interacting with real world. Recognition and categorization of regularity is the first step towards capturing the essential structure of a problem, while at the same time minimizing computational redundancy. One of our research aims under the topic of computational symmetry is to fulfill this gap.
This talk will start with one of our recent results on near-regular texture synthesis and manipulation to demonstrate the advantage of explicitly acknowledging the existence of regularity without losing sight of randomness. The mathematical framework for our work is based on crystallographic groups. I shall introduce our computational models for periodic and bilateral pattern perception, and report their applications in 1) texture synthesis and manipulation, 2) gait analysis, 3) neuroimage analysis and 4) human face, gender and expression classification. Though algorithmic treatment of near-regular pattern analysis shows promising results, I shall also point out several potential perils along the way.
See other events that are part of MIT Machine Vision Colloquium Spring 2004
See other events happening in March 2004
