作者投稿和查稿 主编审稿 专家审稿 编委审稿 远程编辑

计算机工程 ›› 2021, Vol. 47 ›› Issue (4): 262-267. doi: 10.19678/j.issn.1000-3428.0056775

• 图形图像处理 • 上一篇    下一篇

基于小尺度核卷积的人脸表情识别

冯杨, 刘蓉, 鲁甜   

  1. 华中师范大学 物理科学与技术学院, 武汉 430079
  • 收稿日期:2019-12-02 修回日期:2020-02-17 发布日期:2020-03-05
  • 作者简介:冯杨(1994-),女,硕士研究生,主研方向为智能信息处理、模式识别;刘蓉,副教授、博士;鲁甜,硕士研究生。
  • 基金资助:
    国家社会科学基金(19BTQ005);国家科技支撑计划(2015BAK33B00)。

Facial Expression Recognition Based on Small-Scale Kernel Convolution

FENG Yang, LIU Rong, LU Tian   

  1. College of Physical Science and Technology, Central China Normal University, Wuhan 430079, China
  • Received:2019-12-02 Revised:2020-02-17 Published:2020-03-05

摘要: 针对现有表情识别方法中网络泛化能力差以及网络参数多导致计算量大的问题,提出一种利用小尺度核卷积的人脸表情识别方法。采用多层小尺度核卷积块代替大卷积核减少参数量,结合最大池化层提取面部表情图像特征,利用Softmax分类器对不同表情进行分类,并在相同感受野下增加网络深度避免特征丢失。实验结果表明,与FER2013 record、DNNRL等方法相比,该方法的人脸表情识别率更高,能有效实现人脸表情的准确分类。

关键词: 小尺度核卷积, 人脸表情识别, 自然人机交互, 表情特征, 表情分类

Abstract: To address poor generalization ability of network and heavy computation caused by the large number of network parameters in existing expression recognition methods,this paper proposes a facial expression recognition method based on small-scale kernel convolution.The multi-layer small-scale kernel convolution block is used instead of large convolution kernel to reduce the number of parameters.On this basis,the maximum pooling layer is used to extract facial expression image features,and the Softmax classifier is used to classify different expressions.The network depth is also increased under the same receptive field to avoid feature loss.Experimental results show that compared with FER2013 record,DNNRL and other methods,the proposed method has higher recognition rate of facial expression,and can effectively achieve the accurate classification of facial expressions.

Key words: small-scale kernel convolution, facial expression recognition, natural human-computer interaction, expression feature, expression classification

中图分类号: