基于自正则化非负编码和自适应距离度量学习的视频人脸识别  

VIDEO FACE RECOGNITION BASED ON SRNNC AND ADAPTIVE DISTANCE METRIC LEARNING

作　　者：曾贤灏[1] 李向伟[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：《计算机应用与软件》2015年第12期118-122,149,共6页Computer Applications and Software

基　　金：甘肃省教育厅科研项目(2013A-124);甘肃省自然科学基金项目(1107RJZA170)

摘　　要：针对鲁棒人脸识别中K近邻分类无法利用图像集中附加信息的问题,提出基于自正则化非负编码(SRNNC)和自适应距离度量学习的方法。首先,利用样本图像和从样本获得的仿射包模型联合表示一幅图像并进行自正则化非负编码;然后,通过保留大间距架构中数据之间的相似关系来学习更具判别性的马氏距离度量;最后,利用维度加权马氏距离和K近邻分类器完成人脸分类。针对灰度像素值和局部二值模式进行测试,在UCSD/Honda、CMU Moby和You Tube明星数据集上的实验表明,相比其他几种较为新颖的识别方法,该方法取得了更好的识别性能。For the problem that in robust face recognition the k-nearest neighbour classification fails to exploit the additional information in image sets,we proposed an image set face recognition method,it is based on self-regularised nonnegative coding（ SRNNC） and adaptive distance metric learning. First,it uses a sample image and an affine packet model derived from the sample to jointly present an image and carries out the self-regularised nonnegative coding. Then,it learns the more discriminative Mahalanobis distance metric by preserving the similarity relationships among data in a large margin framework. Finally,it uses the dimension weighted Mahalanobis distance and k-nearest neighbour classifier to complete face classification. Test has been made for gray pixel values and local binary patterns. Experiments on UCSD /Honda,CMU Moby and You Tube star data sets showed that the proposed method achieved better recognition performance than several other state-of-the-art methods.

关 键 词：自正则化非负编码 自适应 距离度量学习 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]