Bayesian contextual classification of noise-contaminated multi-variate images
Abstract
A Bayesian contextual classification scheme is presented in connection with modified M-estimates and a discrete Markov random field model. Due to information noise caused by the spatial distribution of training samples and system noise caused by the sensor, modified M-estimates and a preprocessing method of restoring degraded images are implemented to yield spectral models insensitive to the noise sources. The spatial dependency of adjacent class labels is characterized based on local transition probabilities in order to use contextual information. The classification performance which relies on the signal-to-noise ratio is enhanced by removing system noise, and restored images are incorporated in the suggested contextual decision rule. The experimental results show that the suggested scheme outperforms conventional non-contextual classifiers as well as contextual classifiers which are based on the least squares estimates or other spatial interaction models.
Degree
Ph.D.
Advisors
Swain, Purdue University.
Subject Area
Electrical engineering
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.