结合LBP和SVM的视频表情识别方法  

Video expression recognition method combining LBP and SVM

作　　者：姚丽莎[1] 徐国明 房波 何世雄 周欢 YAO Lisha;XU Guoming;FANG Bo;HE Shixiong;ZHOU Huan(Institute of Information and Software,Anhui Xinhua University,Hefei 230088,China)

机构地区：[1]安徽新华学院信息系统软件研究所,安徽合肥230088

出　　处：《山东理工大学学报（自然科学版）》2020年第4期67-72,共6页Journal of Shandong University of Technology:Natural Science Edition

基　　金：安徽省教育厅科技项目(KJ2018A0587);安徽省质量工程建设项目(2018jyssf111);安徽新华学院校级重点科研项目(2018zr006)。

摘　　要：为了提高视频表情实时分类的识别率和实时性,提出LBP特征结合SVM进行决策表情分类的方法。首先获取视频流中的图像并进行预处理,然后使用LBP算子检测人脸,通过多级级联回归树模型对人脸68个关键点进行训练,分别记录表情特征,最后利用SVM训练表情识别模型并预测表情。实验采用Helen dataset作为训练集,CK+数据库作为测试集,平均识别率达到了86.2%,实时性也达到了平均20帧/s。实验结果表明,该方法性能优越,提高了算法的识别率和鲁棒性,同时保证了算法的实时性。In order to improve the real-time recognition rate and real-time performance of video expression classification, LBP(Local Binary Pattern) feature combined with SVM(Support Vector Machine) for decision expression classification is proposed. Firstly, the image in the video stream is captured and preprocessed. Then, the LBP operator is used to detect the face. The 68 key points of the face are trained by the multi-level cascaded regression tree model. The facial expression features of each facial expression are recorded respectively. Finally, it uses SVM to train facial expression recognition models and predict facial expressions. The experiment uses Helen dataset as training set and CK + database as test set. The average recognition rate reaches 86.2% and the real-time performance reaches an average of 20 frames/s. The experiment shows that this method has superior performance, improves the recognition rate and robustness of the algorithm, and ensures the real-time performance of the algorithm.

关 键 词：表情识别 视频 LBP SVM 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]