MARKOV RANDOM MESH MODELS FOR CONTEXTUAL INFORMATION ANALYSIS OF REMOTE SENSING IMAGE

There exist the spectral information and the spatial distribution and structure information in a remote sensing image.To discriminate a pixel we should consider not only its spectral value but also its spatial structure and relation between the pixel and other pixels,that is,the contextual information.This paper is about the remote sensing image analysis using spatial contextual information.Three Markov random mesh models are presented,and the computational methods for the models are given and proved.A Least Square Estimate for correlation parameters of the models is discussed.Using these models we take the joint probability function as the decision function,and use recursive iteration to gradually improve the knowledge of the neighboring classes,and therefore improve the contextual classification results.Experimental results show the effectiveness of the models presented.