Unsupervised multiscale classification using wavelet-domain hidden Markov tree model

A new texture image segmentation algorithm, HMTseg, was recently proposed and applied successfully to supervised segmentation. In this paper, we extend the HMTseg algorithm to unsupervised multiscale segmentation. A Gaussian mixture density is applied to approximate each wavelet coefficient´s joint statistics through modeling each scale and each subband´s histogram. A multiscale Expectation Maximization (EM) algorithm is used to integrate the parameter estimation and classification into one. Then, by the Hybrid Contextual Labeling Tree (HCL T), a Bayesian interscale and intrascale fusion algorithm is applied to raw segmentation results to obtain accurate and reliable final segmentations.