基于卷积神经网络的表情识别  

Expression Recognition Based on Convolutional Neural Networks

作　　者：薛娇 郑津津[1] 

机构地区：[1]中国科学技术大学精密机械与精密仪器系,安徽合肥230026

出　　处：《工业控制计算机》2020年第4期49-50,54,共3页Industrial Control Computer

摘　　要：在传统静态表情识别研究基础上,提出一种简单的人脸裁剪方法,再用浅层卷积神经网络进一步提取特征并进行表情识别。以CK+和JAFFE为实验数据集,进行预处理效果对比实验、数据增强实验、单种表情识别实验和跨数据集六分类实验。结果表明,针对数据量较少的情况,提出的表情识别方法效果明显且鲁棒性更优。Based on the study of traditional static expression recognition,a simple face clipping method is proposed,and then the shallow convolution neural network is used to extract features and recognize expression.Taking CK+and Jaffe as experimental datasets,the experiments of preprocessing,data enhancement,single expression recognition and six classification of cross datasets are carried out.The results show that the proposed method is more effective and robust in the case of less data.

关 键 词：表情识别 人脸裁剪 卷积神经网络 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] TP183[自动化与计算机技术—计算机科学与技术]