基于局部人脸图像的ICA人脸识别方法

ICA face recognition method based on partial face image

提出了一种基于局部人脸图像独立分量分析的特征提取方法。该方法将人脸图像分成若干个相等的部分,将分成的局部人脸图像矩阵作为训练样本,并先后从水平方向、垂直方向提取训练样本的独立分量。相较于传统的独立分量分析(ICA)方法,该方法具有如下优点:有效解决了传统ICA在进行特征抽取过程中的高维小样本问题;将局部人脸图像作为训练样本,这不仅增加了训练样本数,而且有利于提取人脸局部特征;依次从训练样本的水平方向、垂直方向提取训练样本特征,使得提取的特征不仅维数更小,而且能更有效地反映样本的局部信息。以上优点使得提出的算法较传统方法在人脸识别方面更稳定,识别率更高,在Yale人脸库和AR人脸库上验证了该算法的有效性。

A feature extraction method based on partial face image independent component analysis is proposed.This method partitions the facial image into blocks firstly,then takes partial face image matrix as the training sample,and extracts horizontal and vertical independent components from the training samples sequentially.Compared with traditional independent component analysis（ICA）,the developed method has the following advantages： it can effectively solve the problem of ICA in the process of feature extraction,i.e.the number of available training sample is great less than that of the training vector;it takes partial face image as the training sample which doesn’t only increase the number of samples,but also is helpful to get the local features;the method extracts features horizontally and vertically sequentially which makes the features more smaller and reflect the local information of samples more effectively.The advantages referred above let the method more stable and more powerful in face recognition,experiments on the Yale and AR databases validate the effectiveness of the proposed method.