一种快速稀疏分解图像去噪新方法  

A Fast Image Denoising Method Based on Space Decomposition

作　　者：李恒建[1] 张家树[1] 陈怀新[2] 

机构地区：[1]西南交通大学信号与信息处理四川省重点实验室,成都610031 [2]中国电子科技集团公司第10研究所,成都610036

出　　处：《光子学报》2009年第11期3009-3015,共7页Acta Photonica Sinica

基　　金：四川省青年基金(03ZQ026-033);四川省应用基础研究项目(2006J13-10);教育部新世纪优秀人才支持计划项目(NECT-05-0794)资助

摘　　要：提出了一种基于分层树型结构正交匹配追踪算法的快速图像去噪方法.通过选择高斯函数和墨西哥草帽小波母函数构建混合冗余字典,采用分层树状结构表示字典,结合构正交匹配追踪算法,实现图像稀疏表示,提高了图像表示的稀疏性,降低了算法的复杂度.依据噪音能量阈值,通过多次迭代达到图像去噪的目的.实验结果表明,在相同的噪音水平下,该迭代去噪算法取得了较高的较好的PSNR,获得更好的视觉效果.A fast and efficient image denoising method is proposed based on Orthogonal matching pursuit which exploits the tree structure to reduce the complexity of the decomposition.Hierarchical clustering allows for building trees,where each node corresponds to molecule that encompasses the characteristics of all its relative children.Sparse decomposition can be fast solved by Tree based orthogonal matching pursuit,which leads to improved convergence.And,it guarantees the sparsity of results and exactly of reconstructed image.When denoising,the dictionary is built on the Gaussian function which suits well the task of capturing the low-frequency parts of nature image and the anisotropic function which has good ability to capture edges in images.The image denoising approach is taken based on mixture dictionary has the comparison performance with the state-of-the-art under the same noise level both visual quality and PSNR.

关 键 词：图像处理 图像去噪 冗余字典 分层树型结构 正交匹配追踪 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]