Fuzzy Weighted Entropy Attention Deep Learning Method for Expression Recognition

Abstract

Deep learning network has been widely used in facial expression recognition. However, due to the complexity and variability of expression images and the influence of various factors such as illumination and individual differences, the recognition effect of existing methods needs to be improved. In order to improve the expressive ability of deep learning network, an attention mechanism based on fuzzy weighted entropy is introduced into deep learning network and applied to multi-channel facial expression recognition. Firstly, the convolutional neural network model is constructed to extract the convolutional layer multi-channel features of facial image. Then, the fuzzy weighted entropy of different channel features is calculated as the attention weight. Finally, the attention weight and multi-channel features were fused to extract the high-dimensional features of the facial expression image and sent to the full connection layer to complete the facial expression classification. Experiments in CK+ and JAFFE databases show that the average recognition rates of this method obtained 94.86% and 96.88%. It is better than the mainstream methods in recent years.