Monday, November 29, 1999
4:00 PM (refreshments 3:45)
Edgerton Hall, Room 34-101
EECS Colloquium
Abstract
I will describe a new statistical model of images which can be used to perform a number of "high level" computer vision tasks -- such as object recognition and visual texture synthesis. Statistical approaches are alluring because they provide a unified view of learning, classification and generation. To date however, a generic, efficient and unified statistical model for natural images has not appeared. We define a new generative model for the co-occurrence of multiscale coefficients which makes no initial assumption about distributions. The generative process is hierarchical, conditioning fine scale details on coarser scales.
Models of this type can capture the complex statistical structure in images such as a "Persian rug", the spots of a leopard, or a human face. In computer graphics, this model can be used to generate new "texture" images which are almost indistinguishable from the originals. In the area of computer vision, the model can be used to recognize textures, detect faces and other objects.
|
Modified: Nov 23, 1999|
Current events|
Your comments
and inquiries are welcome.