Research On Facial Expression Recognition Based On Attention Mechanism

Facial expressions are an important way for humans to express their emotions and needs.In recent years,facial expression recognition technology has received extensive attention from academia and industry,and it has rich application value in scenarios such as human-computer interaction,safe driving,and online education.With the development of deep learning and neural networks,scholars have proposed a large number of excellent expression recognition models,and these models have achieved good results on related datasets.At present,the research on facial expression recognition has gradually entered the next stage,and the source of expression data used has gradually changed from an undisturbed experimental environment to a complex real environment.However,in the real environment,the collected images usually have interference factors such as occlusion and lack of illumination,so the existing methods are not effective in recognition.In recent years,the attention mechanism has been gradually applied to computer vision tasks.This mechanism enables the network to focus on more important areas and effectively improves the performance of the network.This paper conducts research on facial expression recognition tasks based on the attention mechanism.The main work and contributions are as follows:1.For the task of static image expression recognition,this paper proposes a new convolutional network model.The model is based on the residual network,which introduces local binary pattern features,a regional attention module and a regional weight loss.In this paper,the effectiveness of each module and the good performance of the network are proved by relevant ablation experiments and comparative experiments.Finally,the proposed network achieved correct rates of 60.54%,98.67%,72.25% and 89.11% on the Affect Net dataset,CK+ dataset,FER2013 dataset and RAF-DB dataset,respectively.2.For the task of video frame sequence expression recognition,few people introduce attention mechanism into the model.This paper makes a preliminary attempt and proposes a new convolutional network model.The model introduces a regional differential attention module and a regional self attention module,where the former makes the network focus on regions with large changes,and the latter makes the network focus on regions with richer expression information.Finally,the proposed network achieves 99.69% and 51.96% accuracy on the CK+ dataset and the AFEW dataset,respectively.3.Due to problems such as uneven sample distribution and labeling errors in mainstream expression datasets,this paper constructs a new dataset with relatively balanced sample distribution and better image quality based on CK+,Affect Net and RAF-DB datasets.At the same time,this paper uses the front-end and back-end technologies and the trained network model to realize a simple expression recognition system,which has two functions: image expression recognition and video expression recognition.