小波域隐Markov树模型降噪算法的改进  

An Improved Denoising Algorithm Based on Wavelet-domain Hidden Markov Tree Model

作　　者：石宏理[1] 蔡远利[1] 邱祖廉[1] 

机构地区：[1]西安交通大学电信学院,西安710049

出　　处：《工程数学学报》2005年第3期449-455,共7页Chinese Journal of Engineering Mathematics

摘　　要：提出了适应于更一般情况的基于小波域隐Markov树(HMT)模型信号降噪的改进算法。通常算法以数个零均值高斯函数加权之和描述信号小波系数统计分布,在有些情况下可能带来严重失真。改进算法以一般统计模型描述信号小波系数分布,以最优估计方式对其降噪,能够避免不适当统计模型可能带来的潜在失真。数据仿真表叫该算法有时可以减少降噪信号的MSE(均方误差)80%以上。Proposed is an improved signal-denoising approach that is applicable for more general situations by using wavelet-domain Hidden Markov Tree (HMT) models. In current algorithms, serious distortion may be brought when the distribution of signal wavelet coefficients is described by weighted sum of zero-mean Gaussian functions in some situations. In the improved approach, the wavelet coefficients are denoised by optimal estimation, which is described by a general statistical model. It can effectively avoid the potential distortion caused by inappropriate statistical model in describing the distribution of signal wavelet coefficients. The results of numerical simulation illustrate that the MSE of denoised signal can be reduced by more than 80% with the proposed approach in some situations.

关 键 词：小波 隐Markov树(HMT) 信号降噪 最优估计 

分 类 号：TB535[理学—物理]