融合局部方向模式和卷积神经网络的人脸识别  

Face recognition method with LDP and CNN

作　　者：周鑫燚[1] 张军朝[2] 甘胜江[1] ZHOU Xin-yi;ZHANG Jun-chao;GAN Sheng-jiang(College of Computer Science,Chengdu Normal University,Chengdu 611130,China;College of Electrical and Power Engineering,Taiyuan University of Technology,Taiyuan 030024,China)

机构地区：[1]成都师范学院计算机科学系,四川成都611130 [2]太原理工大学电气与动力工程学院,山西太原030024

出　　处：《计算机工程与设计》2018年第10期3272-3277,共6页Computer Engineering and Design

基　　金：四川省科技厅应用基础基金项目(2015JY0071);成都师范学院高层次引进人才专项科研基金项目(YJRC2014-9);成都师范学院自然科学类培育基金项目(CS14ZD02);山西省重大专项基金项目(20131101029)

摘　　要：为解决光照、姿态、表情、遮挡等变化引起的人脸识别率下降问题,提出一种融合局部方向模式和卷积神经网络的人脸识别方法。采用主动形状模型定位面部区域的关键点位置,校正面部姿态,去除干扰区域,规范面部图像尺寸;对规范化的面部图像进行局部方向模式变换,降低光照影响;在局部方向模式图像上提取五官部位的局部方向特征,在约简局部方向模式图像上采用卷积神经网络提取全局特征,融合构成多模式人脸特征;采用随机森林方法对特征进行学习和分类,实现人脸识别。在ORL、AR和YALE-B数据集上的人脸识别实验结果表明,提出方法的识别率高,对光照、姿态、表情和遮挡变化的鲁棒性强。To solve the problem of face recognition rate degradation caused by variety of luminance,posture,expression and occlusion,a face recognition method fusing local direction pattern(LDP)and convolutional neural networks(CNN)was proposed.Active shape model was used to locate key points of facial region,facial posture was corrected,interference regions were removed,and the size of facial image was normed.Local direction pattern transformation was executed on normalized facial image for reducing luminance influence.Local direction features were extracted from local direction pattern image and global features were extracted from reduced local direction pattern image using convolutional neural networks,and multi-model features were obtained by fusing these features.Random forest was used to learn and classify these features,to realize face recognition.Expe-rimental results of face recognition on ORL,AR and YALE-B datasets show that,the proposed method has high recognition rate,and strong robustness to variety of luminance,posture,expression and occlusion.

关 键 词：人脸识别 局部方向模式 卷积神经网络 随机森林 主动形状模型 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]