由双树复数小波变换的父系数及邻域系数实现图像去噪  

Dual-tree complex wavelet image denoising based on parental and neighboring coefficients

作　　者：卢刚[1] 闫敬文[2] 寇业泉[1] 张建中[1] 

机构地区：[1]厦门大学通信工程系,福建厦门361005 [2]汕头大学电子工程系,广东汕头515063

出　　处：《光学精密工程》2009年第4期916-922,共7页Optics and Precision Engineering

基　　金：航空科学基金资助项目(No.05F07001);国家自然科学基金资助项目(No.60472081)

摘　　要：考虑二维双树复数小波变换(DTCWT)具有良好的平移不变性和方向选择性,基于当前系数与父系数及邻域系数间的关系,构造了DTCWT图像去噪阈值计算公式,提出了一种去噪方法,PNDTCWT。该方法在对图像进行二维DTCWT变换后,利用阈值公式,根据当前系数和父系数及相邻系数计算收缩阈值,对当前系数进行去噪处理。最后,经过二维DTCWT反变换,得到去噪结果。实验结果表明,PNDTCWT的噪声抑制效果明显优于各种基于DWT的去噪方法和其他DTCWT去噪方法。与基于父系数的DTCWT去噪方法相比,PNDTCW的峰值信噪比(PSNR)平均提高了0.5 dB左右。从视觉效果来看,PNDTCW能在去噪的同时较好地保留图像细节,物体轮廓显得比较平滑,不存在传统DWT算法中的混淆现象。By considering the advantages of the 2D Dual Tree Complex Wavelet Transfer（DTCWT） in shift invariance and directionality,a threshold denoising formula based on parental and neighboring coefficients is constituted and a novel Parental and Neighboring DTCWT（PNDTCWT） image denoising method is presented. By proposed method,the shrinkage threshold of each coefficient is calculated to use in denoising for the current coefficient. After 2D DTCWT transfer to an original image, the final image is obtained by the inverse DTCWT of these denoised coefficients. Experimental results show that the denoising performance of the PNDTCWT is better than those of other denoising methods based on DWT or other DTCWT methods,and its Peak Signal Noise Ratios（PSNRs） have improves by 0.5 dB averagely as compared with that of parental coefficients based DTCWT denoising method. In terms of visual quality, PNDTCWT can get the images with more details, smooth profiles and without confusion effect.

关 键 词：双树复数小波变换 父系数 邻域系数 阈值去噪 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]