计算机工程与应用 ›› 2019, Vol. 55 ›› Issue (5): 192-196.DOI: 10.3778/j.issn.1002-8331.1802-0128

• 图形图像处理 • 上一篇    下一篇

多信息融合的深度学习人脸表情识别算法研究

阮  凯,邱卫根   

  1. 广东工业大学 计算机学院,广州 510006
  • 出版日期:2019-03-01 发布日期:2019-03-06

Study on Facial Expression Recognition Algorithm of Multi-Information Fusion Based on Deep Learning

RUAN Kai, QIU Weigen   

  1. School of Computers, Guangdong University of Technology, Guangzhou 510006, China
  • Online:2019-03-01 Published:2019-03-06

摘要: 人脸表情识别作为人机交互系统的重要组成部分,在安防监控、人机交互等领域有广泛的应用,是计算机视觉的研究热点。传统的卷积神经网络方法一般提取单张人脸图像或者人脸标记点作为特征提取的输入数据,未能考虑到人脸全域的表情信息。提出了一种基于三通道多信息融合的深度学习人脸表情识别模型,以人脸图像表情平静到高峰时期标记点坐标的相对位移为输入,提取整个人脸表情图像特征信息,模型融合了稀疏自编码器以提高对边缘特征提取效率。该模型在CK+数据集上进行了训练和测试,实验结果表明,与该领域中的同类算法相比,该算法模型提高了表情识别的准确率。

关键词: 表情识别, 深度学习, 稀疏自编码器, 多信息融合

Abstract: As an important part of human-computer interaction system, facial expression recognition has been widely used in security monitoring, human-computer interaction and other fields, and has always been a research hotspot of computer vision. The traditional convolution neural network method usually extracts single face image or face marker as input of feature extraction and fails to consider the facial expression information in the whole face domain. In this paper, a deep learning facial expression recognition model based on three-channel multi-information fusion is proposed. The feature information of the whole facial expression image is extracted by taking the relative displacement of the facial expression calm to the mark point coordinate in the peak period as the input. The model combines sparse autoencoder to improve the efficiency of edge feature extraction. The experimental results show that compared with similar algorithms in this field, the algorithm model of this paper improves the accuracy of expression recognition.

Key words: facial expression recognition, deep learning, sparse autoencoder, multi-information fusion