Texture image segmentation using multiscale wavelet-domain Hidden Markov Model

We introduce a new image texture segmentation algorithm, based on wavelets and the hidden Markov tree model. Hidden Markov tree model provides a good classifier for distinguishing between textures. We use clustering for determining of texture number in an image to be segmented and we perform raw segmentation for finding the patterns for training the hidden Markov tree model. Using the inherent tree structure of the wavelet coefficients and likelihood computation, we perform texture classification at different scales, then fuse these multiscale classification results using a Bayesian approach to obtain final segmentations. We demonstrate the performance of the algorithm with texture images segmentation.