表情识别算法研究进展与性能比较  

Research and Performance Comparison of Facial Expression Recognition Algorithm

作　　者：崔凤焦 

机构地区：[1]北京科技大学计算机与通信工程学院,北京100083

出　　处：《计算机技术与发展》2018年第2期145-149,共5页Computer Technology and Development

基　　金：国家自然基金重点项目(61432004)

摘　　要：表情是人类情感在面部的表达方式,包含了诸多有用的人类情感和心理活动信息,而表情识别则是研究分析这些信息并进行正确分类的工作。目前,表情识别已成为互联网及相关行业的关注热点,在新兴的智能家居、情感机器人等方面具有较好的应用前景。为此,在分析已有研究成果的基础上,基于Cohn-Kanade库表情,选用卷积神经网络、支持向量机和Adaboost三类算法作为研究对象,通过算法结构设计和参数优化分别得到三类算法相对较优的算法结构,并根据识别过程和识别结果进行了三类算法的对比分析。实验结果及其分析表明,卷积神经网络对Cohn-Kanade表情库的识别效果最好,而Adaboost的处理时间最短,支持向量机的识别效果介于两者之间;表情识别算法的研究及其性能分析为人脸表情识别的实际应用提供了有益的借鉴与参考。As a expression of human emotions in face,emotion contains a lot of useful information about human emotions and mental activity. Emotional recognition is to be analysis of the information and classify them correctly. Presently,emotion recognition has become a newfocus of Internet and related industries,which has a good application prospect in emerging smart home,emotional robots and others. Therefore,on the basis of analyzing existing research,the convolutional neural network,support vector machine and Adaboost are taken as objectbased on Cohn-Kanade expression library. The optimal structures of each algorithm are obtained respectively through design of algorithmstructure and parameter optimization and a comparison on them is made according to the recognition process and recognition result. Experiment shows that convolutional neural network has the best recognition result,while Adaboost has the minimal processing time and supportvector machines somewhere in between. Research and performance analysis of facial expression recognition algorithm provides a reference forthe application of facial expression recognition.

关 键 词：人脸表情识别 卷积神经网络 支持向量机 ADABOOST 人机交互 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]