一种基于半监督稀疏保持投影的人脸识别新方法  

A new face recognition method based on semi-supervised sparisity preserving projection algorithm

作　　者：卫雨彤 林克正[2] 

机构地区：[1]黑龙江大学国际文化教育学院,哈尔滨150080 [2]哈尔滨理工大学计算机科学与技术学院,哈尔滨150080

出　　处：《黑龙江大学自然科学学报》2017年第4期499-504,共6页Journal of Natural Science of Heilongjiang University

基　　金：国家自然科学基金资助项目(61501147);黑龙江省教育厅科学技术基金资助项目(11551087)

摘　　要：针对现实生活中同时存在的无标签样本与标签样本的问题,将稀疏保持投影算法拓展到半监督框架上,提出了一种新的半监督稀疏保持投影算法,实现所有标签样本的充分利用,以保持标签样本给投影子空间的判别能力,建立样本的空间分布信息,提高了人脸识别的精确度。通过(Carngie Mellon University,CMU)人脸数据集对半监督稀疏保持投影算法的有效性进行了验证,结果表明该算法具有较好的人脸识别效果,降低了人脸识别的时间复杂度,提高了人脸识别的精度。Aiming at the problem of the unlabeled samples and labeled samples existing simultaneously in real life, sparsity preserving projection algorithm is extended to semi-supervised framework to propose a new semi-supervised sparsity preserving projection algorithm, so as to make full use of all the labeled samples and keep the determining ability of projection subspace from the labeled samples. The spatial distribution information of space is established to improve the accuracy of face recognition. The CMU face data set is used to validate the effectiveness of the semi-supervised sparsity preserving projection algo- rithm. Experiment results show that the proposed semi-supervised sparsity preserving projection algorithm can achieve better face recognition, reduce the time complexity of face recognition, and improve speed of face recognition.

关 键 词：人脸识别 维数约简 半监督稀疏保持投影 识别精度 数据分类 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]