摘要
针对如今常用的卷积神经网络对人脸表情图片的特征提取不足、关键区域的特征无法精确提取等问题,文章利用不同表情时人脸关键点的变化,并将深度学习方法与聚类方法相结合运用于人脸表情识别中,提出一种基于结构化深度聚类网络(SDCN)的人脸表情识别算法.该网络由GCN图卷积神经网络、K-最近邻(KNN)图构建网络、编码器网络构成.为更好地捕捉到人脸关键点之间的关系和表情信息,利用GCN网络对人脸表情图像中的关键点进行特征提取.该网络输入数据为图结构数据,将人脸关键点数据输入对应的KNN图构建网络以得到人脸关键点的图结构数据.该网络在Fer2013、CK+与JAFFE三个人脸表情数据库上进行实验,获得了较为不错的识别率,在一定程度上证实了算法的有效性.
In response to the problems of insufficient feature extraction of facial expression images and the inability to accurately extract features of key regions using commonly used convolutional neural networks,this paper focuses on the changes in facial key points when different facial expressions are present.By combining deep learning methods with clustering methods,a facial expression recognition algorithm based on Structured Deep Clustering Network(SDCN)is proposed for facial expression recognition.This network consists of GCN graph convolutional neural network,K-nearest neighbor(KNN)graph construction network,and encoder network.To better capture the relationship and expression information between facial key points,this paper uses GCN graph convolutional neural network to extract features of key points in facial expression images.The input data of this network is graph structure data.This article constructs a network by inputting facial key point data into the corresponding KNN graph to obtain the graph structure data of facial key points.The network was simulated on three facial expression databases,Fer2013,CK+,and JAFFE,and achieved good recognition rates,which to some extent confirmed the effectiveness of the algorithm.
作者
胡宇晨
李秋生
HU Yuchen;LI Qiusheng(Research Center of Intelligent Control Engineering Technology,Gannan Normal University,Ganzhou 341000,China;School of Physics and Electronic Information,Gannan Normal University,Ganzhou 341000,China)
出处
《赣南师范大学学报》
2023年第6期56-63,共8页
Journal of Gannan Normal University
基金
江西省教育厅科学技术研究项目(GJJ201408)
江西省研究生创新专项资金资助项目(YC2022-s939)。
关键词
人脸表情识别
结构化深度聚类网络
KNN图构建
图卷积神经网络
人脸关键点
facial expression recognition
structured deep clustering network
construction of KNN diagram
graph convolutional neural network
face key points
作者简介
胡宇晨(1998-),男,江西新余人,赣南师范大学物理与电子信息学院硕士研究生,研究方向:智能信息处理;通讯作者:李秋生(1976-),男,江西南康人,赣南师范大学物理与电子信息学院教授,博士,硕士生导师,研究方向:智能信息处理、目标识别与跟踪等.
