基于改进ANMM及Trace Ratio的人脸识别算法

Face Recognition Algorithm Based on Modified Average Neighborhood Margin Maximizing and Trace Ratio

针对传统ANMM算法计算率低的问题,提出一种基于改进ANMM及Trace Ratio的人脸识别算法.在近邻保持投影的基础上,算法减少了同类数据间由于线性投影而带来的重构误差,同时,保留了类内相似度图的拉普拉斯矩阵的完整性.另一方面,通过构造一个与类内相似度图对应的类外代价图,算法还可以最大化两者间的边界.

In order to resolve the low rate of calculation of traditional ANMM algorithm, a ANMM based on improved and Trace Ratio of face recognition algorithms. Ways to keep the neighbors at the basis of projection, the algorithm reduced because of similar data between the linear projection of the reconstruction error arising at the same time, retained the category similarity graph Laplacian matrix integrity. On the other hand, by constructing a category of ＂similarity＂ graph corresponding categories of the ＂price＂ chart, the algorithm can also maximize the border between the two.