Computer Engineering and Applications ›› 2020, Vol. 56 ›› Issue (23): 161-166.DOI: 10.3778/j.issn.1002-8331.2005-0426

Previous Articles     Next Articles

Attention Hierarchical Bilinear Pooling Residual Network for Expression Recognition

ZHANG Aimei, XU Yang   

  1. College of Big Data and Information Engineering, Guizhou University, Guiyang 550025, China
  • Online:2020-12-01 Published:2020-11-30

注意力分层双线性池化残差网络的表情识别

张爱梅,徐杨   

  1. 贵州大学 大数据与信息工程学院,贵阳 550025

Abstract:

Since facial expression images have subtle inter-class different information and intra-class public information, extracting discriminative local features becomes a key issue. Therefore, an Attention Hierarchical Bilinear Pooling Residual Network(AHBPRN) is proposed. The model uses an effective channel attention mechanism to explicitly model the importance of each channel. It assigns different weights to the output feature map and locates the significant regions according to the weight value. A new hierarchical bilinear pooling layer is added, which integrates multiple cross-layer bilinear features to capture the inter-layer part feature relations, and spatial pooling is carried out in the feature map in an end-to-end deep learning way, so that the proposed network model is more suitable for fine facial expression classification. Experiments on the designed network are conducted on FER-2013 and CK+datasets, respectively, and the highest recognition rates are 73.84% and 98.79%, respectively, which achieves competitive classification accuracy and is suitable for subtle facial expression image recognition tasks.

Key words: facial expression recognition, deep learning, hierarchical bilinear pooling, attention mechanism, ResNet-50

摘要:

由于人脸表情图像具有细微的类间差异信息和类内公有信息,提取具有判别性的局部特征成为关键问题,为此提出了一种注意力分层双线性池化残差网络。该模型采用有效的通道注意力机制显式地建模各通道的重要程度,为输出特征图分配不同的权重,按权重值大小定位显著区域。并添加了一个新的分层双线性池化层,集成多个跨层双线性特征来捕获层间部分特征关系,以端到端的深度学习方式在特征图中进行空间池化,使所提网络模型更适合精细的面部表情分类。分别在FER-2013和CK+数据集上对设计的网络进行实验,最高识别率分别为73.84%和98.79%,达到了具有竞争性的分类准确率,适用于细微的面部表情图像识别任务。

关键词: 面部表情识别, 深度学习, 分层双线性池化, 注意力机制, ResNet-50