摘要
针对面部表情识别在许多领域的重要应用,本文提出了一种基于卷积神经网络(convolutional neural network,CNN)集成的面部表情识别(facial expression recognition,FER)方法。采用3种网络结构不同的卷积神经网络进行训练,利用这些深层模型,使用基于验证准确性的多数投票、简单平均和加权平均的集合方法,在CK+数据集和FER-2013数据集上分别测试单一网络模型和集合网络模型。测试结果表明,单一模型的最佳识别率分别为98.99%和66.45%,集合网络的最佳识别率分别达到99.33%和67.98%,说明使用集合方法的模型比单一模型表现更佳,其中加权平均的集合方法优于简单平均和多数投票,说明本文所提出的方法能够满足面部表情识别的要求。该研究具有一定的实际应用价值。
Aiming at the important application of facial expression recognition in many fields,this paper proposes a Facial Expression Recognition(FER)method based on Convolutional Neural Network(CNN)ensemble.We use three convolutional neural networks with different network structures for training.By these deep models,we use the method of majority voting,simple average and weighted average based on verification accuracy to test the single network model and ensemble network model on the CK+ database and FER-2013 database.The experimental results show that the best recognition rate of the single model is98.99%and 66.45%,and the best recognition rate of the ensemble network is 99.33% and 67.98%,respectively.It shows that the model using the ensemble method performs better than the single model,and the ensemble method of weighted average is better than simple average and majority voting.The method proposed in this paper can meet the requirements of facial expression recognition.This research has certain practical application value.
作者
陆嘉慧
张树美
赵俊莉
LU Jiahui;ZHANG Shumei;ZHAO Junli(School of Data Science and Software Engineering,Qingdao University,Qingdao 266071,China)
出处
《青岛大学学报（工程技术版）》
CAS
2020年第2期24-29,42,共7页
Journal of Qingdao University(Engineering & Technology Edition)
基金
中国博士后科学基金资助(2017M622137)
国家自然科学基金资助(61702293)
教育部虚拟现实应用工程研究中心基金资助(MEOBNUEVRA201601)。
关键词
表情识别
卷积神经网络
网络集合
表情数据集
expression recognition
convolutional neutral network
network ensemble
expression databases
