基于正交梯度差局部方向模式的人脸识别算法  

Face Recognition Algorithm Based on Orthogonal Gradient Difference Local Directional Pattern

作　　者：杨恢先 刘建 张孟娟 曾金芳 Yang Huixian;Liu Jian;Zhang Mengjuan;Zeng Jinfang(School of Physics and Optoelectronics, Xiangtan University, Xiangtan, Hunan 411105, China)

机构地区：[1]湘潭大学物理与光电工程学院,湖南湘潭411105

出　　处：《激光与光电子学进展》2018年第4期144-150,共7页Laser & Optoelectronics Progress

基　　金：湘潭大学校级科研项目(16XZX02);湘潭大学博士启动基金(15QDZ28)

摘　　要：针对差值局部方向模式(DLDP)识别效果不理想的问题,提出一种基于正交梯度差局部方向模式(OGDLDP)的人脸识别方法。分别将3pixel×3pixel和5pixel×5pixel领域的像素灰度值与两组不同的8个Kirsch算子卷积,将两组边缘响应值按照对应编号相互作差并取绝对值,得到8个水平和垂直方向的边缘响应差值;若将3pixel×3pixel领域得到的近邻边缘响应值,按逆时针方向前后作差并取绝对值,同样会得到8个水平和垂直方向的边缘响应差值。取两组边缘响应差值的最大值对应的方向下标,组成一个二位八进制数,形成OGDLDP码。在YALE和AR人脸库进行实验的结果表明:所提算法提高了识别率,且对光照、表情和遮挡变化有较好的稳健性。To solve the problem of poor recognition effect of difference local directional pattern （DLDP）, a face recognition method based on orthogonal gradient difference local directional pattern （OGDLDP） is presented. The pixel gray values in the fields of 3 pixel × 3 pixel and 5 pixel× 5 pixel are convolved with two different sets of 8 Kirsch operators, respectively. We differentiate the two sets of edge response values according to the corresponding numbers and take the absolute value in accordance with the corresponding number to obtain 8 edge response differences in horizontal and vertical directions. If the edge response values near the edge obtained in the 3 pixel× 3 pixel field are made the differences before and after the counterclockwise direction and the absolute values are taken, the 8 edge response differences in horizontal and vertical directions will also be obtained. The OGDLDP code is produced by taking the maximum edge response difference values from the two orthogonal groups corresponding to the direction of the subscript, which are encoded into a double-digit octal number. The experimental results on Yale and AR face databases show that the proposed algorithm improves the recognition rate, and has robustness to the changes of illumination, expression, and shelter.

关 键 词：图像处理 人脸识别 正交梯度差局部方向模式 差值局部方向模式 KIRSCH算子 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]