改进的Curvelet变换图像降噪方法

Noise removal approach using Curvelet transform

与小波变换相比,Curvelet变换等多尺度几何分析方法,可以更好地逼近含线奇异的高维函数。基于Curvelet变换的图像去噪方法,即WindowShrink算法,考虑到Curvelet变换系数之间的相关性,利用软阈值方法降噪。通过窗口邻域操作,对变换后的每一个Curvelet系数自适应地进行萎缩处理,降低噪声系数权重以提高信噪比。实验表明,该方法一定程度上改进了传统Curvelet去噪方法“过扼杀”Curvelet变换系数的缺点,可以较好地保持图像边缘。在噪声方差σ=25时,小波,Curvelet以及WindowShrink去噪算法的峰值信噪比(PSNR)分别为28.59、29.25和29.93,后者明显优于前二者。

Edges in image feature one kind of linear singularity. The aim of Multiscale Geometric Analysis （MGA） which includes Curvelet transform and Ridgelet Transform is to find a kind of optimal representation of such type of image in the sense of nonlinear approximation. Based on Curvelet transform, one image denoising approach is proposed in the paper. Due to the inherent relativity between Curvelet coefficients, this method adopts Curvelet coefficients adaptively with WindowShrink scheme via window/neighborhood processing. Experiments show that this method not only keeps the edge of image but also yields de-noised images with higher PSNR value （PSNR = 29.93 with noise variance σ =25） and better visual quality.