|
Activity Number:
|
137
|
|
Type:
|
Topic Contributed
|
|
Date/Time:
|
Monday, August 12, 2002 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
Section on Statistical Computing*
|
| Abstract - #300967 |
|
Title:
|
Statistical Modeling of Image Sketch
|
|
Author(s):
|
Yingnian Wu*+ and Song Chun Zhu and Cheng-en Guo
|
|
Affiliation(s):
|
University of California, Los Angeles and Ohio State University and Ohio State University
|
|
Address:
|
8130 Math Sciences Building, Los Angeles, California, 90095, USA
|
|
Keywords:
|
vision modeling ; sparse coding ; Markov random fields
|
|
Abstract:
|
The sketch of an image can be defined as a symbolic representation of the image in general, and a line drawing version of the image in particular. Recent results on sparse coding and independent component analysis suggest that human vision first represents a visual image by a linear superposition of a relatively small number of elongate and oriented image bases. With such a representation, a sketch of an image can be obtained by replacing each linear base by a linelet of the same length and orientation. The next step after obtaining such a symbolic representation is to understand the spatial arrangements of the linelets. For this purpose, we propose a statistical model for the sketches of natural texture images. The model is a causal Markov chain model, whose conditional distributions are characterized by a set of simple geometric feature statistics automatically selected from a pre-defined vocabulary. Experiments suggest that this model is capable of capturing important spatial features of a variety of sketch patterns.
|
