一种改进的协同表示算法在人脸识别中的应用

An Improved Collaborative Representation Algorithm in Face Recognition

协同表示算法在人脸识别过程中,受人脸的光照、表情、姿态、特别是训练样本过少等因素影响造成其特征的模糊或者丢失.训练样本数量较少时该问题更为突出,利用原始样本生成虚拟样本再权值融合的协同表示算法虽然对该问题有所改善,却不能大幅度提升人脸识别率.针对以上问题,提出一种改进的协同表示算法,首先生成镜像样本和轴对称样本,以增加单样本人脸特征信息;再在协同表示分类器下分类,对虚拟样本与原始样本加权融合,利用改进的算法增强人脸的特征,并重构误差进行分类识别;最后利用Yale、ORL和FERET人脸数据库实验,比较不同算法识别率.结果表明,在单样本识别中,改进的协同算法识别率比CRC提高2%~17%,比原始样本与对称样本融合提高2%~8%,比原始样本与镜像样本融合提高2%~3%,有效地提高了人脸识别率.

In face recognition,the cooperative representation algorithm is affected by face illumination,expression,posture,and especially the lack of training samples.In practical applications,the number of training samples is often less.The original samples are used to generate a collaborative representation algorithm of virtual sample re-weight fusion,but it cannot greatly increase the face recognition rate.To solve the above problems,a collaborative representation algorithm to enhance the correlation texture information was proposed.First,image samples and axisymmetric samples wre generated to enhance the texture information associated with virtual samples.Then,under the cooperative representation classifier,the virtual sample was weighted and fused with the original sample,the improved algorithm was used to enhance the features of the face,and the error was reconstructed for classification and recognition.Finally,the weighted fusion of virtual samples and original samples was done by using Yale,ORL and FERET databases.The experimental results show that the recognition rate of the algorithm is increased by 2%~17%in single sample recognition,which is 2%~8%higher than the fusion of the original sample and the symmetrical sample,and 2%~3%higher than the fusion of the original sample and the mirror sample.Experimental results show that the algorithm can effectively improve the face recognition rate.