Abstract
This paper is concerned with a method of partitioning a piece of image into some subimages with the same statistical properties. To begin with, assuming the image to be a realization from Markov random field, and using the definition of two-dimensional, discrete Markovian field by J. W. Woods, we introduced the probability density function describing the subimages. And employing the Kullback's divergence measuring the distance between these subimages and transforming above distance, we derived the measure of similarity to all pair of subimages. Measure of similarity derived above, however, does not satisfy the equivalence relation, especially, transitive properties. Therefore we established the equivalence relation by carrying out the composition of a fuzzy relation. Thus we can apply this relation to the partition of image. An example is presented for the illustration of this method.
