Wavelet-Domain Hidden Markov Tree (HMT) Model-Based Image Denoising Algorithm by CHOI from Rice University

This paper by CHOI from Rice University presents an image denoising algorithm based on Wavelet-Domain Hidden Markov Tree (HMT) models, recognized as a classical approach in the field. For researchers focusing on HMT models, this work serves as a highly valuable reference. The author not only introduces the core algorithm but also provides comprehensive explanations of its underlying principles and theoretical background, including wavelet coefficient modeling using Gaussian mixture distributions and context-dependent statistical dependencies through tree-structured transitions. Implementation aspects cover expectation-maximization (EM) algorithms for parameter estimation and Bayesian inference for noise suppression. Experimental results and comparative analyses demonstrate the algorithm's effectiveness and performance advantages in image denoising applications, showcasing quantitative metrics like PSNR improvements and visual quality assessments. Overall, this paper delivers profound theoretical guidance and practical implementation strategies for HMT model researchers and image processing practitioners, establishing itself as a significant contribution to the research literature.