采用双字典协作稀疏表示的光照及表情顽健人脸识别

Illumination and expression robust face recognition using collaboration of double-dictionary's sparse representation-based classification

提出一种采用小波变换(WT)及双字典协作稀疏表示分类(CSRC)的人脸识别方法——WT-CSRC。WT-CSRC首先利用PCA(主成分分析)将小波分解后的人脸高频细节子图融合成高频细节图像;然后用PCA分别对人脸低频图像和高频细节图像进行特征提取,构造低频和高频特征空间,并用训练样本在两种特征空间上的投影集构造低频字典和高频字典;最后将测试样本在两种字典上进行稀疏表示,并引入互相关系数以增强人脸识别的可靠性,实现了人脸的协作分类。实验结果表明,提出的方法提高了人脸识别率,对光照变化及表情变化具有较强的顽健性,并且具有较高的时间效率。

A face recognition method named WT-CSRC was proposed by using wavelet transform（WT） and a collaboration of double-dictionary ＇s sparse representation-based classification（CSRC）. Firstly, the proposed method used principal component analysis（PCA） to achieve the fusion of three high-frequency detail sub-images which were generated by WT, and a integrated high-frequency detail image could be obtained; then, features extracted from the low-frequency images and high-frequency detail images by PCA were used to construct the low-frequency feature space and high-frequency detail space; and low-frequency dictionary and high-frequency dictionary could be constructed by samples＇ projection on two kinds of feature space. Finally, face images could be classified by a collaborative classification via sparse representation in two dictionaries, and the reliability of the recognition could be enhanced by using the cross correlation coefficient. Experimental results show that, the proposed method has high recognition rate with strong illumination and expression robustness with acceptable time efficiency.