面向方向选择的差值局部方向模式的人脸识别  

Face Recognition Based on Direction-Selected Difference Local Direction Pattern

作　　者：顼聪[1] 朱毅[1] 陶永鹏[1] Xu Cong;Zhu Yi;Tao Yongpeng(College of Software,Dalian University of Foreign Language,Dalian 116044,China)

机构地区：[1]大连外国语大学软件学院,辽宁大连116044

出　　处：《南京师大学报（自然科学版）》2020年第4期113-118,共6页Journal of Nanjing Normal University(Natural Science Edition)

基　　金：辽宁省自然科学基金指导项目(2019-ZD-0514);大连外国语大学科研基金项目(2018XJYB29)。

摘　　要：针对目前人脸识别方法中的特征提取缺乏细节和运算量较大的问题,提出一种面向方向选择的差值局部方向模式人脸识别算法(Direction-Selected Difference Local Direction Pattern)DSDLDP,首先利用Kirsch算子计算像素的卷积值,并进行第一次相邻差值计算,然后选择特定方向进行二次差值计算生成DSDLDP模式编码,并利用等价模式降低编码模式种类.最后人脸图像被划分成多个通过DSDLDP编码的图像块,生成对应的直方图,串联起来表示人脸向量.实验结果表明,与当前主流的人脸识别算法相比,DSDLDP算法提取人脸特征更为细致,识别率更高,抗噪声有更好的鲁棒性.Aiming at the problems of lack of detail and large amount of computation in feature extraction in current face recognition methods,a direction-selected difference local direction pattern was proposed.This method uses Kirsch operator to calculate the convolution value of the pixel,and performs the first adjacent difference calculation,and then selects the specific direction for the second difference calculation to generate DSDLDP mode encoding,and uses the equivalent mode to reduce the type of encoding mode.Finally,the face image is divided into multiple image blocks encoded by DSDLDP to generate corresponding histograms,which are connected in series to represent the face vector.Simulation experiment results show that compared with the current mainstream face recognition algorithms,DSDLDP algorithm extracts facial features more meticulously,with higher recognition rate,and has better robustness against noise.

关 键 词：人脸识别 特征提取 KIRSCH算子 局部方向模式 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]