人脸识别中基于熵的局部保持特征提取算法

摘要

Using the variance contribution as the evaluation criteria for feature extraction quality, the traditional local linear manifold feature extraction algorithm cannot guarantee the classification performance after dimension reduction. Thus, a locality-maintain feature extraction algorithm based on Shannon entropy is proposed. For the proposed algorithm, the classification uncertainty of the feature extraction is described by the overall Shannon entropy that is also regarded as the evaluation criteria of feature extraction. The analysis and face recognition experiments show that,compared with the local linear manifold feature extraction algorithm, the classification performance of feature extraction can be improved with the proposed algorithm while keeping the data local features.%传统的局部线性流形特征提取算法以方差贡献率为特征提取质量评价准则,不能保证降维后的分类性能.为此,提出了一种基于香农熵的局部保持特征提取算法,采用总体熵描述特征提取对分类的不确定性,并作为特征提取的评价准则.分析与人脸识别实验表明,相对于局部线性流形特征提取算法,提出方法在保持数据局部特性的同时,改善了特征提取的分类性能.