采用结构自适应窗的非局部均值图像去噪算法  

A Non-Local Means Algorithm for Image Denoising Using Structure Adaptive Window

作　　者：郝红侠[1,2] 刘芳[1,2] 焦李成[2] 武杰[1,2] 

机构地区：[1]西安电子科技大学计算机学院,西安710071 [2]智能信息感知与图像理解教育部重点实验室,西安710071

出　　处：《西安交通大学学报》2013年第12期71-76,共6页Journal of Xi'an Jiaotong University

基　　金：国家重点基础研究发展计划资助项目(2013CB329402);中央高校基本科研业务费专项资金资助项目(K5051203002;K5051203007);国家自然科学基金资助项目(61072106;61173090)

摘　　要：针对图像去噪处理中的非局部均值(NLM)算法相似性度量结果不够准确的问题,提出了一种采用结构自适应窗的非局部均值图像去噪(SAW-NLM)算法。首先利用从含噪图像中提取的初始素描图将含噪图像划分为结构区和非结构区,然后对这两部分区域分别采用基于结构方向的自适应窗和各向同性窗来搜索相似图像块,最后利用这些相似图像块得到当前待估计像素的去噪结果。为了抑制伪纹理现象,在估计过程中采用了块估计的方式。自适应窗有效结合了图像的结构方向和灰度信息,因此能够更准确度量图像块的相似性。实验结果表明:SAW-NLM算法具有更优的边缘保持和平滑效果,与传统NLM算法相比,峰值信噪比最大可提高1.1dB,图像结构相似度也提高了4.6。A structure adaptive window based non-local means （SAW-NLM） algorithm for image denoising is proposed to solve the problem that the results of the similarity measurement are not accurate in the non-local means （NLM） method. The proposed algorithm divides a noisy image into two parts, structural area and non-structural area, by using the primal sketch map that is extracted from the noisy image. Then, similar samples are respectively searched in these two parts using structure direction based adaptive window and isotropic window. Finally, the denoising result is estimated from these samples. Moreover, the window-wise method is employed in the estimation of the interest to suppress the pseudo texture phenomenon. Since the structure direction and the gray information are jointly used in the adaptive window, the similarity can be more accurately measured. Experimental results show that the SAW-NLM algorithm has advantages in the edge preservation and smooth effects. Comparisons with the traditional NLM algorithm show that SAW-NLM method improves the peak signal-to-noise by 1.1 dB and the structural similarity index by 4.6.

关 键 词：图像去噪 度量 结构自适应窗 估计 非局部均值 

分 类 号：TN911.73[电子电信—通信与信息系统]