基于优化卷积神经网络的人脸图片微表情识别方法研究  

Research on Facial Image Micro Expression Recognition Method Based on Optimized Convolution Neural Network

作　　者：张爱民 ZHANG Ai-min(Party School of Guoyang County Party Committee,Guoyang 233600,China)

机构地区：[1]中共涡阳县委党校,安徽涡阳233600

出　　处：《内蒙古民族大学学报（自然科学版）》2022年第1期37-42,共6页Journal of Inner Mongolia Minzu University：Natural Sciences

摘　　要：常规微表情识别方法,采用计算单演相位编码的方式,提取人脸图片微表情特征效果不佳,导致识别的人脸图片微表情存在严重的微表情识别混淆问题,为此提出基于优化卷积神经网络的人脸图片微表情识别方法研究。采用齐剪裁、图像序列插帧和帧数归一化3种方式,预处理人脸图片;量化单演相位值,计算单演相位、方向和幅值编码,提取人脸图片微表情特征;选择2种非线性函数激活网络,通过前向传播和反向训练,识别人脸图片微表情。实验结果表明:研究方法较此次实验选择的2组方法,JAFFE数据库微表情混淆种类分别减少9种和7种;Cohn-Kanade数据库微表情识别混淆分别减少8种和5种,减少人脸图片微表情识别出现的混淆问题种类。The conventional micro expression recognition method,which uses the method of calculating single acting phase coding,has poor effect in extracting the micro expression features of face images,resulting in serious confusion in micro expression recognition. Therefore,a face image micro expression recognition method based on optimized convolution neural network is proposed. Face images are preprocessed by three methods:uniform clipping,image sequence frame interpolation and frame number normalization;The single performance phase value is quantified and the single performance phase,direction and amplitude encoding are calculated and the micro expression features of face images are extracted;Two nonlinear functions are selected to activate the network and the micro expression of face image is recognized through forward propagation and back training. The experimental results show that compared with the two groups of methods selected in this experiment,the types of micro expression confusion in JAFFE database are reduced by 9 and 7 respectively;The confusion of micro expression recognition in Cohn Kanade database is reduced by 8 and 5 respectively and the types of confusion problems in face image micro expression recognition are reduced.

关 键 词：优化卷积神经网络 人脸图片 微表情特征 表情识别 混淆问题 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]