基于参考人脸表情的多特征融合表情识别算法  

Multi-feature fusion expression recognition algorithm based on referenced facial expression

作　　者：王菽裕 宋俊芳[1] 王蒙 吴思 关琛夕 WANG Shuyu;SONG Junfang;WANG Meng;WU Si;GUAN Chenxi(School of Information Engineering,Xizang Minzu University,Xianyang 712082,China)

机构地区：[1]西藏民族大学信息工程学院,陕西咸阳712082

出　　处：《现代电子技术》2021年第7期77-81,共5页Modern Electronics Technique

基　　金：西藏自然科学基金(XZ2017ZRG-53(Z));校内重大培育(19MDZ03);陕西省教育厅专项(19JK0887);大学生创新创业训练项目(S201910695060)。

摘　　要：人脸表情识别具有巨大的发展潜力,可用于智能人机交互、大众娱乐、安全驾驶、辅助医疗和在线教育等应用场景。文中提出一种新的表情识别算法。首先,通过局部二值模式(LBP)、局部矢量模式(LVP)和Gabor局部矢量模式(GLVP)创建参考人脸表情,从参考人脸表情提取各类表情特征;其次,采用布谷鸟搜索算法将提取的特征进行聚类;最后,在测试阶段,利用极限学习机(ELM)依据不同表情类别的不同特征进行识别;并对所提出的算法和现有K最近邻(KNN)分类算法、支持向量机(SVM)方法,从识别精度和误差等角度进行了分析比较,所提算法均取得了良好的实验效果。The face expression recognition has tremendous development potential and can be applied in situations such as intelligent human-computer interaction,popular entertainment,safe driving,assistant medical treatment and online education.A new expression recognition algorithm is proposed in this paper.In the algorithm,local binary pattern(LBP),local vector pattern(LVP)and Gabor local vector pattern(GLVP)are used to create reference face expressions,from which various expression features are extracted.The Cuckoo search algorithm is used to cluster the extracted features.In the test stage,the extreme learning machine(ELM)is used to recognize different features of different expressions.In addition,the proposed algorithm and existing algorithms,for instance,K-nearest neighbor(KNN)classification algorithm and SVM method,are analyzed and contrasted in terms of recognition accuracy and error.It verifies that the experimental results obtained with the proposed algorithm are better than that obtained with the existing algorithms.

关 键 词：表情识别 参考人脸表情 表情特征 特征聚类 LBP LVP GLVP ELM 

分 类 号：TN911.73-34[电子电信—通信与信息系统] TP391[电子电信—信息与通信工程]