WIAMIS 2006, 7th International Workshop on Image Analysis for Multimedia Interactive Services, April 19-21, 2006, Incheon, Korea
In this paper, we investigate some properties of a new type of 2D Hidden Markov Model, based on the notion of Dependency Tree. DT-HMMs avoid the complexity of regular 2D HMMs by changing the double horizontal and vertical spatial dependencies into a random uni-directional dependency, either horizontal or vertical. We explore various issues about the effect of this random choice. This type of probabilistic model can be useful in many applications for image and video segmentation, classification, and others.
Type:
Conférence
City:
Incheon
Date:
2006-04-21
Department:
Data Science
Eurecom Ref:
1969
See also:
