摘要
针对人脸识别准确率易受人脸表情变化影响的问题,提出了一种分区域特征提取的人脸识别算法。首先,在预处理图像上标记出14个人脸关键点,并将人脸图像划分为表情易变区域和不变区域;然后,分别用Gabor+LBP和Gabor+分块LBP两种特征提取通道对表情不变区域和易变区域进行特征提取;最后,将所得的特征直方图级联,并进行身份验证。经FERET(face recognition technology),LFW(labled faces in the wild)及自制人脸数据库验证,文中算法准确率分别达到了99.14%,98.5%,96.52%。在FERET数据库中,该文算法准确率较DeepID和Gabor+分块LBP算法分别提高了1.88%和3.6%,F1(调和平均数)分别提高了1.8%和2.86%。实验结果表明,分区域特征提取的人脸识别算法对人脸表情变化具有很强的鲁棒性。
In order to solve the problem that the accuracy of face recognition is easily affected by the change of facial expression,a face recognition algorithm based on feature extraction by region is proposed.Firstly,on the processing image 14 key points of people face is marked,and the face image is divided into expression variable region and invariable region.And then respectively using Gabor+LBP and Gabor+block LBP two kinds of feature extraction channel for expression invariant region and mutable region to extract feature.Finally,the feature histogram is cascaded and the authentication is carried out.Verified by FERET(face recognition technology),LFW(labled faces in the wild)and self-made Face database,the accuracy reached 99.14%,98.5%and 96.52%,respectively.In the FERET database,the accuracy of the algorithm in this paper is 1.88%and 3.6%higher than DeepID and Gabor+block LBP algorithm,and the F1(harmonic average)is 1.8%and 2.86%higher.Experiments show that the face recognition algorithm based on regional feature extraction is robust to face expression changes.
作者
李云红
聂梦瑄
苏雪平
周小计
何琛
LI Yunhong;NIE Mengxuan;SU Xueping;ZHOU Xiaoji;HE Chen(School of Electronics and Information, Xi′an Polytechnic University, Xi′an 710048, China)
出处
《西北大学学报(自然科学版)》
CAS
CSCD
北大核心
2020年第5期811-818,共8页
Journal of Northwest University(Natural Science Edition)
基金
陕西省科技厅青年科学基金资助项目(2019JQ-255)
西安市科技局高校人才服务企业项目(2019217114GXRC007CG008-GXYD7.2,2019217114GXRC007CG008-GXYD7.8)。
关键词
分区域特征
人脸易变区域
人脸不变区域
特征直方图
subregional features
face variant region
face invariant region
feature histograms