摘要
对影响因素未知且没有先验信息的污染图像进行恢复和重构是图像处理的一项主要任务。ICA(Independent Component Analysis,独立分量分析)是20世纪90年代后期发展起来的一种盲信号处理方法,并成功应用于图像盲分离。近年来ICA技术得到了进一步发展,出现了多种算法。为了分析各种算法在图像盲分离中的优劣,对SOBI(二阶盲辨识)、JADE(联合近似特征矩阵对角化)、FastICA(快速独立分量分析)和KICA(基于非线性子空间的核独立分量分析)算法进行了比较实验。结果表明,KICA算法分离效果最好,FastICA算法次之;但是如果源图像或者混合图像中含有噪声,则以上四种方法分离效果都不佳。
It is a main task to separate the interesting image from the image polluted by Gaussian noise and other uninteresting signals without any prior information.Independent component analysis（ICA） is a new and important method of blind signal processing developed in 1990s,and it is used in blind image separation successfully.Recently,several new algorithms were proposed.In order to analyze which algorithm is better in blind image separation,four algorithms including SOBI（second order blind identification）,JADE（joint approximate diagonalization）,FastICA and KICA（kernel independent component analysis） are used in this blind image separation experiment.The results show that KICA can separate the mixed images with best effect,followed by FastICA.But if the mixed images polluted by Gaussian noise,we cannot get a good result by using any of these four ICA algorithms.
出处
《工程勘察》
CSCD
2012年第4期70-75,共6页
Geotechnical Investigation & Surveying
基金
中南大学自由探索项目学生潜质发掘专项(项目编号:201011200143)
关键词
独立分量分析
图像处理
图像盲分离
independent component analysis（ICA）
image processing
blind images separation
