基于变换域单源点筛选的高效盲图像分离  

An Efficient Blind Image Separation Based on Screening Single Source Points in Transform Domain

作　　者：徐金东[1] 余先川[1] 胡丹[1] 张立保[1] 

机构地区：[1]北京师范大学信息科学与技术学院,北京100875

出　　处：《电子学报》2013年第11期2193-2198,共6页Acta Electronica Sinica

基　　金：国家自然科学基金(No.41272359;No.41072245);教育部博士点基金(No.20120003110032);中央高校基本科研业务专项基金(No.2012LZD05)

摘　　要：在基于稀疏成分分析的盲图像分离中,有效聚类点数直接影响分离的速率和精度,针对此问题,提出一种基于变换域单源点筛选的高效盲图像分离算法.根据变换域单源点的定义及分析,通过比较混合图像的一级Haar小波对角分量与水平分量的绝对方向,筛选出"单源点",有效地约简了参与估计混合矩阵的聚类点数,使信号特征更加稀疏.仿真实验结果表明,Haar小波域的单源点筛选方法能更快、更精确地估计混合矩阵,且统计直方图显示,该方法对潜在变量分析有所启发.In the field of blind image separation based on sparse component analysis ,the separation efficiency and accuracy is directly affected by the valid number of clustering samples .For this problem ,a new algorithm for detection of points in the Haar wavelet domain where only single source contributions occur was proposed .The algorithm identified the single source points （SSPs ） by comparing the absolute direction between diagonal component and horizontal component of Haar wavelet coefficients of mixed images .After screening SSPs ,the signal features are sparser .The experiment results showed that the algorithm could estimate the mixing matrix faster and more accurately ,and it could inspire to identify the latent variables by statistical histogram .

关 键 词：盲图像分离 单源点 稀疏成分分析 潜变量分析 小波变换 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]