基于稀疏分解的置换混叠图像盲分离  

Blind separation for permuted alias images based on sparse decomposition

作　　者：段新涛[1,3] 方勇[1,2] 

机构地区：[1]上海大学通信与信息工程学院特种光纤与光接入网重点实验室,上海200072 [2]新型显示技术与应用集成教育部重点实验室,上海200072 [3]河南师范大学计算机与信息技术学院,新乡453000

出　　处：《高技术通讯》2012年第4期368-373,共6页Chinese High Technology Letters

基　　金：国家自然科学基金（60872114）,上海市科委重点实验室项目（08DZ2231100）,上海市重点学科基金（S30108）和上海市教委科研创新重点项目（09ZZ89）资助.

摘　　要：针对一类存在形态差异的置换混叠图像，提出了一种基于稀疏分解的置换混叠图像盲分离新算法。该算法为了利用置换混叠图像中被置换区域和置换区域在不同的字典上稀疏表示存在的稀疏差异，分别选择Contourlet变换基字典和局部离散余弦变换基字典作为分离的特征域，通过置换混叠图像在两个特征域上的稀疏形态分解，把纹理图像从置换混叠图像中分离出来。实验结果表明，对于一类包含分片光滑和纹理成分的置换混叠图像，该算法能够有效地把纹理图像从置换混叠图像中分离出来，同时，对纹理图像的大小、位置、个数和类型具有鲁棒性。This paper presents a new blind separation algorithm for permuted alias images based on sparse decomposing, aiming at a type of permuted alias images with morphological diversity. The method uses the Contourlet dictionary and the local discrete cosine transform dictionary respectively as the characteristic fields of separation so as to use the sparsity diversity existing in sparse representation of the permuting and permuted regions of a permuted alias im- age. And then the texture image is separated from the permuted alias image by decomposing the permuted alias im- age on the two characteristic fields. The simulation results show that the proposed algorithm can separate effectively texture images from the permuted alias images regardless of size, location number and types of texture image for a permuted alias image comprising the piecewise smooth part and the texture part.

关 键 词：置换混叠图像 稀疏表示 冗余字典 形态差异 稀疏分解 盲分离 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]