结合多特征和跨连通道加权的面部表情识别  

Facial Expression Recognition Combined with Multiple Features and Cross-channel Weighting

作　　者：柳璇 唐颖军 黄淑英 LIU Xuan;TANG Ying-jun;HUANG Shu-ying(School of Software and Internet of Things Engineering,Jiangxi University of Finance and Economics,Nanchang 330032,China)

机构地区：[1]江西财经大学软件与物联网工程学院,南昌330032

出　　处：《小型微型计算机系统》2021年第2期399-404,共6页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(61762043,61562035)资助;江西省自然科学基金项目(20192BAB207022)资助;江西省教育厅科学技术研究项目(GJJ190249,GJJ160425)资助.

摘　　要：面部表情是人类表达情感的主要方式.本文提出一种将手工特征和深度学习特征相结合,以跨连通道加权模块为基础的面部表情识别方法.将灰度图、局部二值模式特征、Sobel特征作为三通道特征输入,以深度可分离卷积代替标准卷积;同时引入跨连通道加权模块,通过建模不同通道特征之间的关系,更加高效地进行不同层次特征的融合.本文方法在CK+和JAFFE两个常用表情数据集上进行了验证,取得了高达99.77%和99.48%的准确率,证明了本文方法的有效性与可行性.Facial expressions are the main way for human to convey emotional information.A method based on cross-channel weighted module combining manual features and deep learning features is proposed to recognize facial expression.The grayscale image,LBP feature,and Sobel feature are used as three-channel feature map,and depthwise separable convolution is used instead of standard convolution.At the same time,the cross-connect channel weighted module is introduced to model the relationship between different channel features to perform more efficiently fusion of different levels of features.The proposed method is verified on the two public benchmarking expression datasets,extended Cohn-Kanade(CK+)and JAFFE,and achieved high accuracy of 99.77%and 99.48%separatelly,which proves the effectiveness and feasibility of the method.

关 键 词：表情识别 跨连网络 通道加权 深度可分离卷积 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]