具有最大散度无关性的局部保持投影算法

Maximum Scatter Difference Uncorrelated Based on Locality Preserving Projections

为解决传统流形学习方法不能有效利用人脸类间信息的问题,提出具有最大散度无关性的局部保持投影算法(Maximum-Scatter-Difference-Uncorrelated Locality Preserving Projections,M ULPP).该算法是最大类间无关性的局部保持映射算法,通过求取一组最优的无关鉴别矢量集,既达到特征映射后的类间散度保持最大、类内散度保持最小,同时又满足最佳鉴别矢量之间具有最大统计不相关性,从而提高算法的识别性能.在AT&T和YALE标准人脸图像库上的实验结果表明,MULPP算法具有较高的识别率.

With the purpose of solving the problem that the traditional manifold learning method cannot effectively using the face between-class information,Multiple Locality Preserving Projections （MULPP） based on Locality Preserving Projections is proposed in this article.The algorithm is a kind of vector discriminant calculation method which has the best between-class independency,and its purpose is to look for the optimal identification vector set which cannot only achieve the result that the feature space has the largest between-class divergence and the smallest intra-class divergence but also satisfy the maximal statistical uncorrelation between the optimal identification vectors after the projection,so as to improve the recognition performance of the algorithm.The results of the experiments on Standard face image database YALE and AT＆T indicate that MULPP has high recognition rate.