一种小波域HMT模型参数初始化方法

摘要：

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

关键词：Wavelet、Hidden Markov tree model、Parameter initialization、EM algorithm

所属期刊栏目：30

分类号：TP3(计算技术、计算机技术)

资助基金：国家自然科学基金60073053;高等学校博士学科点专项科研项目

在线出版日期：2004-01-08（万方平台首次上网日期，不代表论文的发表时间）

页数：共3页

页码：85-86,77