基于小波局部统计特性的图像去噪方法

Image Denoising Algorithm Based on Local Statistical Property in Wavelet Domain

本文提出一种利用图像在小波域上局部统计特性的自适应去噪方法。首先在LMMSE准则下,推导出小波系数在局部区域的恢复公式。为进一步精确地估计理想小波系数的局部方差,本算法提出利用尺度间和子带内的相关性,即利用粗尺度下小波系数的局部方差预测精细尺度下相应位置的小波系数为噪声成分的概率,以及常规估计下的小波系数的局部方差是否小于某个门限值判断其是否为噪声成分。然后以这些局域窗内非噪声成分系数估计理想小波系数局部方差。实验结果表明,本算法与传统算法相比,对图像质量有进一步地改善,尤其是对细节丰富的图像表现地更为突出。

A adaptive image denoising method is presented which utilizes the local statistical property of wavelet coefficients. Firstly, under the rule of LMMSE we deduce the restoration formula of the wavelet coefficients in the local domain. Secondly, in order to improve the estimation accuracy of the ideal wavelet coefficients, the method exploits the correlation of inter-scale and intra-subband. The local variation of a wavelet coefficient in the coarse scale is used to predict whether the coefficient in the next fine scale is noise component ,which exploits the first correlation. As well, by thresholding, the regular estimation of local variance is used to predict whether the coefficient is noise component, which exploits the second correlation. Then, the ideal local variance of a wavelet is estimated from those coefficients which are not the noise components in the local window. The experiments show that, compared with traditional method, the algorithm makes a great improvement of image quality .especially for the image with much detailed information.