Facial expression recognition method based on confrontation elimination

Abstract

The invention relates to a face expression recognition method based on confrontation elimination, and relates to the field of computer vision. The method comprises the following steps: firstly, on the basis of a deep convolutional neural network, constructing a facial expression recognition network, and training the facial expression recognition network through a loss function on a natural human facial expression data set to enable facial expression features to be distinguished more easily; then, an improved adversarial elimination method being utilized to actively eliminate part of key features of an input image, a new data set being generated to train a new network with different weight distributions and feature extraction capabilities, and the network being forced to carry out expression classification discrimination according to more features; the influence of interference factors such as shielding on the network recognition accuracy is reduced, and the robustness of the facial expression recognition network is improved; and finally, obtaining a prediction result of final expression classification by adopting network integration and a relative majority voting method. According to the method, the accuracy of the facial expression recognition network is improved, and the interference of shielding factors on the network is effectively reduced.

Description

technical field [0001] The invention relates to the field of computer vision computing, in particular to a face expression recognition method based on confrontation elimination. Background technique [0002] With the gradual development of deep learning and the continuous expansion of computer vision applications, involuntary facial expression recognition based on laboratory environments is no longer a challenge, and the focus of academic research has shifted to facial expression recognition under natural conditions. . Since the first natural environment facial expression recognition competition EMotiW was held, more and more algorithms and high-quality natural facial expression data sets for natural facial expression recognition have been proposed by researchers. Facial expressions in natural environments will be obviously affected by changes in illumination, occlusion, and the posture of the task itself. Extracting effective facial expression features in natural environme...

AI Technical Summary

Problems solved by technology

Although the natural facial expression dataset is closer to the facial expressions acquired in the real scene, due to the small sample size of the natural facial expression dataset and the presence of interference factors such as skin color, illumination and occlusion, the network overfitting phenomenon is relatively serious. For some key features Unobtrusive images are more prone to misclassification

[0003] At present, the facial expression recognition network based on the attention mechanism has achieved good results on the natural expression dataset, but the human facial expression recognition network based on the attention mechanism needs to provide additional input images artificially, and requires a large number of attention subclasses. The network performs feature extraction on these images. During the training process, the backbone network and the sub-network need to run at the same time, so it takes up a lot of computer resources.

Images