摘要： 局部保持投影(LPP)算法未利用样本类别信息进行人脸识别，提取的特征不适合分类。为解决该问题，提出一种基于排斥图和吸引图的LPP算法。在K近邻图的基础上建立排斥图和吸引图，使排斥图反映2个邻近但不同类样本之间的关系，吸引图反映2个同类但不近邻样本之间的关系，结合两者进行特征提取，定义样本相似性度量，以去除原始特征提取噪声和特征值变异的影响。在Feret和Yale人脸数据库上的实验结果表明，该算法的识别率高于主成分分析算法和传统LPP算法。

关键词: 人脸识别, 局部保持投影, 排斥图, 吸引图, 相似度, 特征提取

Abstract: In order to overcome the shortcoming that Locality Preserving Projection(LPP) algorithm does not use label information for face recognition and the extracted feature of LPP can not achieve good classification results, this paper proposes a LPP based on rejection graph and attraction graph algorithm. The algorithm rejection graph and attraction graph based on K-nearest-neighbor(KNN) graphs. Rejection graph reflects the relationship between two near-by samples which is in different classes and attraction graph reflects the relationship between two samples which is not near-by, but in the same class. It combines rejection graph and attraction graph to extract feature, defines the similarity of samples to remove the effects of noise and eigenvalues variation when extracting the original feature. Experiments on Feret and Yale face image datebase show that the recognition performance of the proposed algorithm is higher than Principal Component Analysis(PCA) algorithm and LPP algorithm.

Key words: face recognition, Locality Preserving Projection(LPP), rejection graph, attraction graph, similarity, feature extraction

中图分类号: 

• TP391.4