基于监督协同近邻保持投影的人脸识别算法

Supervised Collaborative Neighborhood Preserving Projection Based Algorithm for Face Recognition

基于流形学习理论的近邻保持嵌入算法(Neighborhood Preserving Embedding,NPE)能够发现数据集中隐含的内蕴结构,但当训练样本不足时,无法准确发现数据的内在流形结构,从而影响算法的识别效果。针对这一问题,对NPE算法进行改进,提出了监督协同近邻保持投影算法(Supervised Collaborative Neighborhood Preserving Projection,SCNPP)。该算法在类别信息的指导下构建近邻图,使同类样本间的几何关系得到保持,利用协同表示弥补NPE因样本不足造成的表示误差,以一个有效保持样本近邻关系、准确发现数据内在流形结构的权值矩阵计算投影矩阵,提高分类效果。在FERET、AR和Extended Yale B人脸数据集上的实验验证了该算法的有效性。

Neighborhood preserving embedding （NPE） algorithm based on manifold learning theory can discover the intrinsic structure behind data set. But in the scenery of face recognition,algorithm can＇t detect the intrinsic structure accurately due to the insufficient of data, sequentially, the performance of NPE is influenced. In order to solve the problems of NPE in face recognition, a supervised collaborative neighborhood preserving projection （SCNPP） algorithm was presented. The proposed algorithm constructs the neighborhood graph under the guidance of category information, makes the geometric relationship between the same samples be preserved effectually,utilizes the collaborative representation to remedy the representation errors of NPE caused by the lack of data, calculates the projection matrix with a weight matrix which preserves the neighborhood relationship effectually and discoveries the intrinsic manifold structure of data accurately,improves the performance of classification. Extensive experiments on popular face databases （FERET, AR and Extended Yale B） verify the effectiveness of the proposed method.