Abstract: The paper briefly introduced the basic conception of Generalized Gaussian Distribution (GGD) and studied the distributional property of wavelet coefficients, then analyzed the principle of BayesShrink and pointed out the exiting shortcomings. A local adaptive wavelet denoising algorithm was proposed based on the redundant wavelet transform and the relativity among wavelet coefficients in the subband. The new method selected a proper neighboring window by centering the current coefficient within it, and estimated the corresponding ideal standard deviation and threshold for the centered coefficient, and then made shrinkage on it by soft thresholding. The experimental results show the new method effectively filters the noise, reserves more texture and detail of images and gets higher Peak Signal-to-Noise Ratio (PSNR) value and better visual expression.

Key words: image denoising, Generalized Gaussian Distribution （GGD）, minimum Bayes risk, redundant wavelet transform, threshold

摘要： 简要介绍了广义高斯分布的基本概念和小波系数的分布特性，分析了BayesShrink法的原理并指出其存在的不足。以冗余小波变换为基础，利用子带内小波系数之间的相关性，提出了一种局部自适应的图像去噪策略。以当前小波系数为中心，选取尺寸合适的邻域窗口，以该窗口为单位估计相应的理想标准差和局部阈值，再通过软化处理达到系数收缩的目的。实验表明，该算法能有效滤除图像噪声，较好地保留了图像纹理和细节等重要信息，取得了较高的峰值信噪比和较好的视觉效果。

关键词: 图像去噪, 广义高斯分布, 最小贝叶斯风险, 冗余小波变换, 阈值