基于熵检测的图像噪声方差小波域估计  

An Entropy-Based Estimation of Noise Variance in Wavelet Domain

作　　者：李天翼[1,2] 王明辉[1] 常化文[1] 陈淑清[1] 

机构地区：[1]四川大学计算机学院,成都610065 [2]四川大学制造学院,成都610065

出　　处：《北京邮电大学学报》2011年第5期1-5,共5页Journal of Beijing University of Posts and Telecommunications

基　　金：国家自然科学基金项目(61071162)

摘　　要：提出一种在小波域中基于熵值检测的图像噪声方差估计算法.利用小波变换能显著降低图像信号的熵而并不改变高斯噪声熵的特性以及噪声熵值与噪声方差之间呈对数关系变化的规律,定量地分析了含噪图像在小波高频对角子带中系数的熵值随噪声幅值的变化规律,揭示出这种变化关系对图像具有较强的鲁棒性,从而利用这种变化关系,通过对含噪图像小波域熵值的检测对高斯噪声进行估计.仿真结果表明,提出的算法能有效估计出图像中噪声的方差,并且受图像细节影响较小,其性能优于现有其他算法.A new approach with entropy inspection in wavelet domain is proposed for estimation of the variance of Gaussian noise in images. With advantage of wavelet transform, the entropy of image signal degrades notably with the Gaussian noise entropy unchanging. It denotes that the noise entropy value varies with the noise levels in manner of logarithm. The relation between the entropy values of the noisy image wavelet coefficients and the noise levels is traversed in quantitative form. A formula is given to depict this functional relation. Experiment indicates that such a relation is robust to images. A noise estimation is thus made with the entropy inspection of the noisy image in wavelet domain and the established formula as apriori knowledge. Meanwhile, a simulation shows that the proposed approach can achieve more exact value while little influenced by image details.

关 键 词：小波变换 熵 高斯噪声 方差估计 

分 类 号：TN929.53[电子电信—通信与信息系统]