A facial expression recognition method based on generative antagonistic network

Abstract

The invention relates to a facial expression recognition method based on a generative antagonistic network, which relates to computer vision technology. Firstly, a facial expression generation networkbased on generative antagonistic network is designed and pre-trained. The network is composed of a generator and two discriminators, which can generate a face map with random identities of specifiedfacial expressions. Then a facial expression recognition network is designed, which receives the facial expression map from the training set and the random facial expression map generated by the facial expression generation network, and uses an intra-class loss to reduce the facial expression feature differences between the real samples and the generated samples. At the same time, a real sample-oriented gradient updating method is used to promote the feature learning of the generated samples. Finally, according to the trained facial expression recognition network model, the final facial expression recognition results are obtained from the final flexible maximum classification layer of the model.

Description

technical field [0001] The invention relates to computer vision technology, in particular to a human facial expression recognition method based on a generative confrontation network. Background technique [0002] In the past few years, automatic facial expression recognition has attracted the attention of many experts in the field of computer vision. Automatic facial expression recognition technology has important display significance in many application scenarios, such as social robots, healthcare and human-computer interaction. Although the automatic facial expression recognition technology has made good progress over the years, it still faces great challenges, especially in complex environments, such as different poses, lighting and occlusions, etc., the recognition of automatic facial expression recognition rate still needs to be improved. [0003] Existing facial expression recognition techniques can be divided into two categories: methods based on manually designed f...

AI Technical Summary

Problems solved by technology

However, directly using these generated pictures as training data will bring new problems, such as how to generate high-quality face pictures, and how to ensure that these pictures can have a positive impact on training convolutional neural networks.

If these problems are not handled well, it is likely that not only will the performance of the model not be improved, but also the feature representation ability of the model will be weakened.

Images