一种能实现免脱帽人脸识别系统算法  

A System Algorithm for Face Recognition Without Taking off the Hat

作　　者：张彦虎[1,2] 鄢丽娟 张彦军[4] ZHANG Yanhu;YAN Lijuan;ZHANG Yanjun(School of Computer and Information Engineering,Guangdong Songshan Politechnic,Shaoguan 512126,China;Jose Riazal University,Mandaluyong Metro Manila 1550,Philippines;School of Computer Science and Engineering,Sun Yat-Sen University,Guangzhou 510006,China;Gansu Wuhuan Highway Engineering Co.,Ltd.,Lanzhou 730000,China)

机构地区：[1]广东松山职业技术学院计算机与信息工程学院,广东韶关512126 [2]国父大学,菲律宾曼达路尤1550 [3]中山大学数据科学与计算机学院,广州510006 [4]甘肃五环公路工程有限公司,兰州730000

出　　处：《计算机测量与控制》2022年第2期244-251,共8页Computer Measurement &Control

基　　金：广东省普通高校特色创新项目(2019GKTSCX041);广东省高职教育精品课程建设项目(粤教职函[2018]194.50);韶关市科技计划(社会发展与农村科技专项)资金项目(2018SN041)。

摘　　要：针对人脸识别系统在人脸被遮挡情况下识别率低的问题,为进一步提升人脸在遮挡情况下的识别率,文章提出一种通过图像多方向梯度值,使用融合、补偿等方式产生可以对原图像进行特征描述的特征图像,通过对特征图进行一系列处理后实现人脸识别的算法;算法首先计算图像四方位的梯度值;其次对4个梯度值进行融合运算,产生合融梯度、差融梯度;再次以合融梯度、差融梯度作为补偿变量在原图像上进行适当系数的补偿,形成人脸图像特征图;然后对特征图依次进行直方图统计、主成分分析后,使用SVM分类器进行分类识别;使用Matlab2016试验仿真平台在ORL、CMU_PIE等多个人脸数据库上进行测试,分别取得100%、92.21%的准确率,结果表明推荐算法在人脸被遮挡情况下的识别率具有很好的表现。Aimed at the face recognition system for lower recognition rate problem under the obscured condition,in order to further improve human face recognition rate under the condition,a kind of more direction from the image gradient values is put forward,the use of fusion,compensation and other way can describe the characteristics of the image characteristics of the original image,the characteristics of the figure are based on a series of processing to realize the algorithm of face recognition.Firstly,the gradient value of image quadrangle is calculated.Secondly,the fusion operation is performed on the four gradients to generate fusion gradient and differential fusion gradient.Thirdly,the fusion gradient and differential fusion gradient are used to compensate the proper coefficients on the original image to form the face image feature image.Then,after the histogram statistics and principal component are analysised,the SVM classifier is used for classification and recognition.Matlab 2016 experimental simulation platform is used to test on multiple face databases such as ORL and CMU_PIE,and the accuracy reaches 100%and 92.21%respectively.The results show that the recognition rate of the recommendation algorithm has a good performance in the case of blocked faces.

关 键 词：梯度 图像梯度 图像梯度补偿 人脸识别 身份识别 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]