Multiscale segmentation of remotely sensed images using pairwise Markov chains

Abstract

Among the statistical approaches to image modeling, Markov random fields have recently gained significant attention, especially in texture segmentation. Different from Markov random fields, in pairwise Markov chains, the class field is not necessarily a Markov field, an advantage in the segmentation of texture images without any model approximation. Supervised texture segmentation of a multiscale image is introduced in a pairwise Markov chain tree model using the wavelet domain. The essence of this tree-structured probabilistic graph is based on capturing the statistical properties of the wavelet transforms and the intrinsic characters of textural regions of any multispectral image.