摘要
在生活中很多方面都要用到面部表情识别技术,但是人脸不同部位的表情变化非常细微,加上技术的有限性,在之前的研究中很多识别的特征都是人为定义的。本文基于卷积神经网络模型开展表情识别的研究工作,为了尽可能的提高最终表情识别的准确性,需要大量的样本图片训练,优化,所以采用了fer2013数据集用来训练、测试,此数据集由35886张人脸表情图片组成,其中,测试图28708张,公共验证图和私有验证图各3589张,所有图片中共有7种表情。在预处理时把图像归一化为48×48像素,训练的网络结构是基于VGG网络结构基础上改进的自定义的网络结构,正文中会具体介绍,通过不断地改进优化,缩小损失率,最终能达到较准确的识别出人的面部表情的结果。
Facial expression recognition technology is used in many aspects of life,but the expression changes in different parts of the face are very subtle.Coupled with the limitation of technology,many of the recognized features in previous studies are artificially defined.In this paper,based on the convolutional neural network model,the research work of expression recognition is carried out.In order to improve the accuracy of the final expression recognition as much as possible,a large number of sample pictures are needed for training and optimization.Therefore,the fer2013 data set is used for training and testing.It consists of 35886 face expression pictures,including 28708 test pictures,3589 public verification pictures and private verification pictures,and 7 expressions in all pictures.The image is normalized to 48×48 pixels during preprocessing.The trained network structure is based on the improved custom network structure according to the VGG network structure.It will be specifically introduced in the text,and the loss rate is reduced by continuously improving the optimization.In the end,it is possible to achieve a more accurate result of recognizing a persons facial expression.
作者
靳显智
林霏
王叶
JIN Xian-zhi;LIN Fei;WANG Ye(School of Electrical Engineering and Automation,Qilu University of Technology(Shandong Academy of Sciences),Jinan 250353,China)
出处
《齐鲁工业大学学报》
CAS
2021年第3期64-69,共6页
Journal of Qilu University of Technology
基金
国家自然科学基金联合项目(U2006222)
山东省教育厅研究生教研计划创新项目(SDYY16032)。
关键词
卷积神经网络
面部表情识别
分类算法
convolutional neural network facial
expression recognition
classification algorithm
