基于反馈机制的抗混合噪声的盲图像源分离方案  

A scheme of blind sources separation based on feedback mechanism for mixed images with noise

作　　者：徐金东[1,2] 余先川[2] 胡丹[2] 张立保[2] 邢海花[2] 

机构地区：[1]北京师范大学信息科学与技术学院,北京100875 [2]曲阜师范大学物理工程学院,曲阜273165

出　　处：《中国科学：信息科学》2013年第7期920-931,共12页Scientia Sinica(Informationis)

基　　金：国家自然科学基金(批准号:61071103,41272359,11001019);北京市自然科学基金(批准号:4102029);高等学校博士学科点专项科研基金(批准号:20120003110032);山东省优秀中青年科学家科研奖励基金(批准号:BS2010DX012)资助项目

摘　　要：针对基于稀疏成分分析的盲图像源分离算法无法解决混合噪声问题,本文提出一种采用反馈机制的盲源分离算法.通过小波域稀疏成分分析和置零反馈的方法,逐次分离出各支路信号.实验结果表明,该方法无需大量的迭代运算,与传统稀疏成分分析法相比,能有效地分离高斯白噪声参与的混合图像,与经典快速独立成分分析法相比,取得了更高的分离精度.Considering that the algorithm based on sparse component analysis cannot perfectly solve the problem about mixed images with noise, a new scheme utilizing feedback mechanism was presented to separate image sources blindly. Via sparse component analysis in wavelet domain and zero setting feedback mechanism, every channel of signal was extracted step by step. The experiment results showed that the proposed method does not need vast iteration and can separate all sources from the mixtures effectively. Compared to classic FastlCA algorithm, tile preseated algorithm got higher accuracy.

关 键 词：稀疏成分分析 盲源分离 小波变换 反馈 混合噪声 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]