一种优化自正则化非负编码的图像集人脸识别方法  

The Research of Image Sets Face Recognition Based on SRNNC

作　　者：张志勋[1] 曾贤灏[1] 

机构地区：[1]兰州工业学院软件工程学院,甘肃兰州730050

出　　处：《佳木斯大学学报（自然科学版）》2017年第4期620-622,共3页Journal of Jiamusi University：Natural Science Edition

基　　金：甘肃省教育厅科研计划项目(2014A-120);甘肃省教育厅高等学校科研计划项目(1114-03)

摘　　要：针对图像集分类中无法确保特征空间的每个位置上的良好性能的问题,提出一种优化自正则化非负编码方法。首先,将图像集由萁样本图像和从样本获得的仿射包联合表示,并将其集间距离表示为自正则化非负约束的仿射包的最近点间的距离;然后,通过维度加权马氏距离进行度量;最后,利用K近邻分类器完成人脸分类。结果表明:在UCSD/Honda明星数据集上针对灰度像素值模式的实验相比其他几种较为新颖的识别方法具有更好的识别性能。For the issue that simple nearest neighbor classification fails to exploit the additional informa- tion in image sets in robust face recognition, an image set face recognition method based on self - regularized nonnegative coding is proposed. Firstly, the image set represented by using sample images obtained from the sample and from the affine package, and the distance between sets that is the nearest point to affine self normal- ized nonnegative constrained packet distance. Then, measuring by dimension weighted Mahalanobis distance. Finally, using K neighbor classifier to finish classification. The results show that the proposed algorithm has bet- ter recognition performance than the other new methods in the UCSD/Honda star data set.

关 键 词：图像集分类 自正则化非负编码 视频人脸识别 马氏距离 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]