摘要
面部表情识别是地铁、火车站、机场等复杂环境中安检监控的重要任务,通过识别监控图像中行人的面部表情可以筛选出可疑分子。针对因监控图像模糊和面部表情拍摄不全而引起的识别准确率低等问题,提出一种改进的InceptionV4面部表情识别算法,改进InceptionV4的网络结构,使其更好地适应面部表情识别任务。基于深度学习中的Tensorflow平台对面部表情类数据进行训练,在面部表情验证集上进行测试,在输入图像为299×299时,识别准确率高达97.9%,改进后的算法在保证识别精度的同时,降低表情在类内差距较大、图像模糊和面部表情拍摄不全情况下的误识率,提高系统鲁棒性。
In order to solve the problem of low recognition rate of facial expression recognition caused by in⁃tra-class gaps,blurred images and incomplete facial expression shooting in real environment,an improved Inception V4 recognition facial expression was proposed.In view of the difficulty of security inspection in complex environ⁃ments such as subways,railway stations,and airports,the network structure of Inception V4 was improved to make it more suitable for facial expression recognition tasks.This article uses the Tensorflow platform in deep learning to train facial expression data,and tests it on the facial expression verification set.When the input image is 299×299,the recognition accuracy is as high as 97.9%.The improved algorithm guarantees the accuracy of the recognition.At the same time,it reduces the misunderstanding rate in the case of large gaps within the class,blurred images and in⁃complete facial expression shooting,and improves the robustness of the system.
作者
张景异
梁宸
吴攀
陈亮
刘韵婷
Zhang Jingyi;Liang Chen;Wu Pan;Chen Liang;Liu Yunting(ShenyangLigong University,School of Automation and Electrical Engineering,Shenyang 110159,China)
出处
《光电技术应用》
2020年第1期56-63,共8页
Electro-Optic Technology Application
基金
国家重点研发计划（2017YFC0821004）
国家重点研发计划（2017YFC0821001）
辽宁省自然科学基金（20170540788）
辽宁省教育厅基本科研项目(LG201707)
