基于鉴别稀疏邻域保持嵌入算法的人脸识别技术

Face Recognition Based on Discriminant Sparsity Neighborhood Preserving Embedding

本文针对人脸识别技术中的关键问题-特征提取,设计了一种新颖的人脸特征提取算法。结合稀疏表示和邻域保持嵌入算法,提出了一种鉴别稀疏邻域保持嵌入算法(DSNPE)。该算法可以基于稀疏表示创建近邻图,包括类内样本近邻图和类间样本近邻图,分别讨论了类内紧致性和类间稀疏性等问题。同时该鉴别稀疏邻域保持嵌入算法可以较好地利用样本类别信息,故该算法具有监督性。基于最大间距准则,建立了鉴别稀疏邻域保持嵌入算法的目标函数,并描述了该算法的基本流程。最后,在Yale、ORL和AR人脸数据库上进行了相关实验,并与SPP、NPE、LPP、MMC、LDA、PCA等算法比较,实验结果表明:基于鉴别稀疏邻域保持嵌入算法具有更好的人脸识别性能。

As the key issues of face recognition,a novel face extraction algorithm is proposed in this paper. A discriminant sparsity neighborhood preserving embedding algorithm( DSNPE) is proposed based on sparse representation and neighborhood embedding. The algorithm can create neighborhood graph with sparse representation,including within-neighborhood graph and between-neighborhood graph,respectively discusses the problems about within-class compactness and between-class sparseness. At the same time the DSNPE algorithm can take advantage of the sample category information better,so this algorithm has supervision. The objective function of DSNPE algorithm is established based on the maximum distance between standards,and its basic flow is described. Finally,the related experiment is conducted with comparison of SPP,NPE,LPP,MMC,LDA and PCA algorithm on Yale,ORL and AR face databases,the experimental results show that the discriminant sparsity neighborhood preserving Embedding algorithm has better performance of face recognition.