TITLE:     Image Segmentation with Hidden Markov Model

 

SPEAKER:   Johan Lim, Stanford University

 

DATE:      2-3 P.M., Tuesday, January 28, 2003

 

LOCATION:  Sigma, 1L (PA)

 

HOST:      Vinay Deolalikar

 

-----------------------------------------------------------------------

 

ABSTRACT:

Image segmentation (i.e., partitioning an input image into several homogeneous regions) is a crucial preliminary phase in many practical applications. Adopting the model with spatial interactions, several authors recently proposed simultaneous classification and segmentation procedures through which smooth boundaries were obtained. Among these models, the Hidden Markov Model (HMM) is the most popular. Several important statistical issues, which have been often overlooked in practice, arise when segmenting an image using the HMM. We address these issues and discuss the insights they provide for more complex HMMs in other applications. In particular, a theory of estimation in general HMMs will be presented.