Contextual Hidden Markov Tree Model for Signal Denoising

The hidden Markov tree (HMT) model is a novel statistical model for signal and image processing in the wavelet domain. The HMT model captures the interscale persistence property of wavelet coefficients, but includes only a tiny intrascale clustering property of wavelet coefficients. In this paper, we propose the contextual hidden Markov tree (CHMT) model to enhance the clustering property of the HMT model by adding extended coefficients associated with the wavelet coefficients. The extended coefficients are regarded as leaves to link wavelet coefficients in the wavelet tree model without destroying the wavelet persistence property; hence, the training approach of the HMT model can be modified to estimate the parameters of the CHMT model. In experiments, the proposed CHMT model produced better results than the HMT model for signal denoising. Furthermore, the CHMT model needs fewer iterations of training than the HMT model to get the same denoised results.