基于稀疏表示与特征融合的人脸识别方法  

Face recognition method based on sparse representation and feature fusion

作　　者：木立生 吕迎春[1] MU Lisheng;LüYingchun(School of Information Engineering,Taiyuan University of Technology,Jinzhong 030600,China)

机构地区：[1]太原理工大学信息工程学院,山西晋中030600

出　　处：《现代电子技术》2018年第9期83-86,90,共5页Modern Electronics Technique

基　　金：国家自然科学基金资助项目(51279122)~~

摘　　要：针对人脸识别在有遮挡、表情、光照的变化或受到噪声污染时鲁棒性变差问题,提出一种基于稀疏表示与特征融合的人脸识别算法。首先采用低秩恢复算法得到训练样本和测试样本的干净人脸图像,提取干净人脸图像的LBP,HOG,Gabor三种特征向量;然后对部分训练样本进行SRC分类测试,根据SRC的识别结果与分类残差定义一个损失函数,再利用正则化最小二乘法计算出使损失函数最小的权重向量;最后根据该权重向量重构规则化残差进行分类。在ORL,Extended Yale B和AR数据库上进行实验,结果表明,该算法优于利用单一特征识别的方法,并且对光照、噪声、遮挡等因素产生的影响有较好的泛化性能。Since the robustness of face recognition becomes worse due to the changes of sheltering,expression and illumination,or noise pollution,a face recognition algorithm based on sparse representation and feature fusion is proposed.The low-rank recovery algorithm is used to get the clean face images of training samples and test samples,and their feature vectors(LBP,HOG,Gabor)are extracted.The SRC classification test was performed for some training samples.A loss function is defined according to the rec-ognition result and classification residual of the SRC.The regularization least-square method is used to calculate the weight vector with minimum loss function,according to which the regularization residual is reconstructed for classification.The experiment of the method was performed on ORL,Extended Yale B and AR databases.The results show that the algorithm is superior to the single feature recognition method,and has better generalization performance on the influence of illumination,noise and sheltering.

关 键 词：人脸识别 稀疏表示 低秩恢复 特征融合 鲁棒性 泛化性能 

分 类 号：TN911.73-34[电子电信—通信与信息系统] TP391.41[电子电信—信息与通信工程]