摘要
针对传统基于自商图像的方法忽略对特征进行选择的问题,提出了一种结合自商图像和随机投影的人脸识别方法。采用自商图像法对人脸图像进行预处理,削弱光照影响;然后通过线性判别分析构造初始样本空间,利用多次随机投影将样本投影到不同的子空间,从而提取更具完备性和判别性的光照不变特征。最后用最近邻分类器对样本进行分类。在Yale B和AR人脸库上的实验表明:所提算法可以提取对光照鲁棒且具有鉴别性的人脸特征,从而提高光照变化条件下人脸识别的准确率。
Aiming at the problem that the method based on self quotient image ignores the selection of features,a novel face recognition approach based on self quotient image and random projection is proposed.Firstly,the self quotient image method is used for preprocessing face image in order to remove the effect of illumination.Secondly,constructs an initial sample space using linear discriminat analysis.Furthermore,samples are projected to different subspaces by iterative random projection and extracts more completeness and discriminative of the illumination invariant feature.Finally,the nearest neighbor classifier is used to classify sample.Experiments on Yale B and AR face database show that the proposed method can extract face features which are robust to illumination and discriminative,and improve the accuracy of face recognition under illumination change.
作者
曹洁
朱晶晶
李伟
王进花
CAO Jie;ZHU Jingjing;LI Wei;WANG Jinhua(College of computer and communication,Lanzhou University of Technology,Lanzhou730050,China;College of Electrical and Information Engineering,Lanzhou University of Technology,Lanzhou730050,China)
出处
《传感器与微系统》
CSCD
2019年第7期121-124,共4页
Transducer and Microsystem Technologies
基金
国家自然科学基金资助项目(61263031,61763028)
甘肃省自然科学基金资助项目(1506RJZA105)
关键词
人脸识别
光照变化
自商图像
随机投影
face recognition
illumination variation
self quotient image
random projection