|
Activity Number:
|
208
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Monday, August 7, 2006 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
Section on Bayesian Statistical Science
|
| Abstract - #306257 |
|
Title:
|
Video Segmentation Using a Bayesian Online EM Algorithm
|
|
Author(s):
|
Johan Lindström*+ and Finn Lindgren and Kalle Åström and Jan Holst and Ulla Holst
|
|
Companies:
|
Lund University and Lund University and Lund University and Lund University and Lund University
|
|
Address:
|
Box 118, Lund, SE-22100, Sweden
|
|
Keywords:
|
recursive estimation ; mixture models ; video segmentation ; Bayesian models
|
|
Abstract:
|
A method for video segmentation using Bayesian models and an online EM algorithm is presented. The method models each layer as a Gaussian mixture, with local, per pixel, parameters for the background and global parameters for the foreground. The online EM algorithm also uses a progressive learning rate allowing the relative update speed of each Gaussian component to depend on how long that component has been observed and similar foreground components are merged using a Kullback-Leibler distance. Performance of the algorithm for gray-scale and RGB videos as well as on output from a Prewitt edge detector is compared to that of another algorithm. Especially for the edge detector, performance increases dramatically.
|
