行列分块的核独立成分分析的人脸识别

提出一种行列分块的核独立成分分析(RC-KICA)的人脸识别方法。RC-KICA先对人脸图像矩阵按行列分块;然后对训练样本集依次进行行和列的核独立成分分析,得到左右解混矩阵;最后把训练样本子块投影到解混矩阵构成的特征空间进行特征提取及分类识别。RC-KICA更大程度地降低了样本维数,更好地解决了KICA高维小样本的缺陷。在YALE人脸库上的实验结果表明RC-KICA优于KICA和B-KICA。

The ranks of the block and kernel independent components analysis for face recognition

The ranks of the block and kernel independent components analysis (RC-KICA) for face recognition is proposed in this paper. First of all, the face image matrix are divided into blocks by columns and rows according to this method. Then kernel independent components analysis used followed by rows and columns in training sample set to obtain the left-unmixed matrix and right-unmixed matrix. At last, the all blocks are projected to the eigenspace which was constructed by the left-unmixed matrix and right-unmixed matrix to feature extract and .recognition. RC-KICA method reduced the sample dimension greatly. Besides, it solved the defects of small number and high-dimensional samples from KICA method. The experimental results on YALE face database indicate that the performance of RC-KICA is superior of that KICA and B-KICA.