Deep Learning-Based Micro-Expression Recognition Algorithm Research

Abstract

In order to improve the accuracy and speed of micro-expressions, a modified model based on densenet and eca is proposed. Microfacial expression is a brief, weak facial change, its characteristics are similar, dense, difficult to extract and identify, and the improved model can be adapted to the characteristics and location of the interest. In particular, the eca attention module was added after the densenet model, using the densenet network to extract the rich characteristics of micro-expressions, and the eca attention module to recalibrate the feature channel and focus on the more subtle expression changes. In order to verify the validity of this method, the experiment was conducted in the micro-emotive data set, and compared with the resnet network and the densenet network, the experimental results showed that the improved model significantly improved the performance of micro-expression recognition, and had strong generalized ability and robustness.