基于SURF和形状上下文的人脸匹配算法  

Face image matching algorithm based on SURF and shape context

作　　者：师硕[1] 于洋[1] 杨志坚 于明[1] Shi Shuo;Yu Yang;Yang Zhijian;Yu Ming(School of Computer Science&Engineering,Hebei University of Technology,Tianjin 300401,China;North Automatic Control Technology Institute,Taiyuan 030006,China)

机构地区：[1]河北工业大学计算机科学与软件学院,天津300401 [2]北方自动控制技术研究所,太原030006

出　　处：《计算机应用研究》2018年第10期3197-3200,共4页Application Research of Computers

基　　金：天津市科技计划资助项目(14RCGFGX00846);天津市科技计划资助项目(15ZCZDNC00130);河北省自然科学基金面上资助项目(F2015202239)

摘　　要：针对人脸匹配在光照、姿态、表情等背景因素影响下匹配正确率低的问题,提出一种基于SURF和形状上下文(SC)的人脸图像匹配算法。在对图像进行人脸区域检测和重构积分图预处理的基础上,利用两次SURF算法提取人脸特征点并匹配,第一次用SURF进行粗匹配得到初始匹配集,并据此计算待匹配人脸图像间的尺度差和角度差,以此作为约束条件第二次进行SURF精匹配,以获得更多的匹配点对。最后采用不依赖位置信息的形状上下文算法剔除误匹配点对。在IMM和Georgia人脸数据库上与目前流行的人脸匹配方法进行实验对比,实验结果显示该算法有效增加了匹配点对数目,并提高了人脸图像匹配正确率,具有更好的稳定性和鲁棒性。This paper proposed a face image matching algorithm based on SURF and shape context to solve the problem that the matching accuracy was low under the influence of background factors such as illumination,posture and expression.Based on the preprocessing of the face region detection and reconstruction of integral image,this paper used twice SURF to extract face feature points and match.The first rough match obtained the initial match set and calculated the scale difference and the angle difference between the faces to be matched.The SURF fine matching performed again with the scale difference and the angle difference as the constraint condition to obtain more matching pairs.Finally,the paper used the shape context algorithm,which did not rely on the position information,to eliminate the misplaced point pairs.Compared with other currently popular face matching algorithms on IMM and Georgia face databases,the experimental results show that the algorithm improves the matching accuracy on the basis of obtaining more matching points,and has better stability and robustness.

关 键 词：人脸图像匹配 加速鲁棒特征 形状上下文 误匹配点剔除 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]