Multiscale Texture Segmentation Using Hybrid Contextual Labeling Tree

Wavelet-domain hidden Markov tree (HMT) model has been recently proposed and applied to image processing, e.g., image segmentation. A new multiscale image segmentation method, called HMTseg, was proposed by Choi and Baraniuk using the waveletdomain HMT. In this paper, we study the HMTseg algorithm and investigate the Contextual Labeling Tree which is used for the context-based Bayesian interscale fusion of the multiscale classification information. In order to attain more accurate multiscale characterizations with improved segmentation results, we develop three new context structures which have different advantages on the interscale fusion. Then we propose a hybrid context-based interscale fusion algorithm where the three contexts are serially cascaded so that the Bayesian estimation is conducted based on the three contexts respectively and sequentially. The proposed method outperforms the original HMTseg algorithm by improving the accuracies of both texture classification and boundary localization.