摘要
在人脸识别中,人脸图像受到表情、光照、遮挡、姿态变化、特别是训练样本数量的影响,而现实中经常只获得少量的训练样本,由于原始样本生成虚拟样本可以增加训练样本的数量,分析提出原始样本与轴对称样本融合的协同表示算法。首先生成镜像样本与轴对称样本,再在协同表示分类器下分类,最后加权值融合,分析不同权值下的人脸识别率。实验结果显示原始样本、镜像样本与轴对称样本融合能提高识别率,而原始样本与轴对称样本融合的识别率更加优越,较原始样本,识别率提高2%~9%,比原始样本与镜像样本融合高1%~5%。结果表明本文提出方法能有效提高人脸识别率。
For face recognition,the face image are affected by the variations of expression,lighting,occlusion,pose and especially the number of training samples.However,in practical application,we only have insufficient training samples.The collaborative representation algorithm of combining the original training samples with the axial-symmetry samples is proposed because the original training samples generate the corresponding virtual training samples to increase the number of training samples.Firstly,the original training samples generate the corresponding mirror samples and axial-symmetry samples.Secondly,the reconstruction errors are obtained by using collaborative representation based classification.Finally,the variant reconstruction errors are combined with different weighted number to compare face recognition rates.The experimental results show that the face recognition rates are increased by combining the original training samples with the mirror samples and the axialsymmetry samples.The face recognition rates of combining the original training samples with the axialsy-mmetry samples are 2%~9% and 1%~5% better than the original training samples and theoriginal training samples with the mirror samples respectively.It shows that the paper's method is effective in face recognition.
出处
《液晶与显示》
CAS
CSCD
北大核心
2017年第12期987-992,共6页
Chinese Journal of Liquid Crystals and Displays
基金
四川理工学院科研项目(No.2015RC16)~~
关键词
人脸识别
镜像样本
轴对称样本
协同表示
权值融合
face recognition
mirror samples
axial-symmetry samples
collaborative representation
weight fusion
