线性判别分析特征提取稀疏表示人面识别方法-《郑州大学学报(工学版)》

文章信息/Info

Title:: Sparse Representation-based Face RecognitionMethod by LDA Feature Extraction

Author(s):: ZHANG Yong; DANG Lan-xue; Institute of lmage Processing and Pattern Recognition，Henan University，Kaifeng 475004 , China

Keywords:: LDA; sparse representation ; feature extraction ; face recognition

摘要:: 针对稀疏表示分类(SRC)算法采取随机脸法提取的数据特征判别力较弱问题,提出一种线性判别分析特征提取稀疏表示人脸识别方法.该方法首先采用线性判别分析算法求解最优判别投影子空间,然后把训练样本投影到该子空间以提取相应的数据特征,并用训练样本的数据特征做字典来表示测试样本数据特征.更进一步来说就是,通过提取出测试样本稀疏特征的向量,和测试样本的数据特征进行比对找出其联系和差别并表示出比对后的残差.最后根据构造的残差找出样本的类别来实现其识别目的.通过在Extend Yale B和CMU PIE人脸数据库上一系列的测试,证明该方法具有很好的识别效果.

Abstract:: To solve the problem that the features extracted by randomfaces method have weak discriminative a-bility in sparse representation-based classification ( SRC),a sparse representation - based face recognitionmethod by linear discriminant analysis ( LDA) feature extraction was proposed. Firstly,LDA is used to solvethe optimal discriminative projective subspace,and then the training samples are projected onto the subspaceto extract the features of the training samples. Using the features of the trainings samples as the dictionary, thefeatures of the test sample can be sparsely represent as linear combination of the atoms of the dictionary. Fur-thermore,using the sparse coefficients associated with the special class,this method approximates the featuresof the test sample and calculates the reconstruction error between the features of the test sample with its ap-proximation associated with the special class. Based on the reconstruction error associated with special class,the test sample can be classified accurately. Experimental results on Extend Yale B and CMU PIE face data-bases show that face recognition method proposed in this paper has a good performance.