A Parameter Initialization Method for Wavelet-Based HMT Models
一种小波域HMT模型参数初始化方法

Wavelet transformation and hidden Markov model are used in wavelet-based HMT model for analyzing and processing images. Expected Maximization (EM) algorithm used in training model results in slow convergence. The persistence, exponential decay characteristics of wavelet coefficient are analyzed. A model parameter initialization method is proposed. This method provides reasonable initial model value, reduces training time greatly. Its application in image de-noising demonstrates is validity.