Research on a Micro-Expression Recognition Algorithm based on 3D-CNN

Abstract

Micro expression is a kind of natural human expression, which lasts for a short time and is not easy to detect. Due to the subtle spatiotemporal variation of micro-expressions, the recognition of micro-expressions is still a big challenge. Although many scholars have made some attempts in the recognition of micro-expressions, the accuracy of the recognition problem is still not ideal. In order to take advantage of 3D convolution, we propose an improved model of micro expression recognition based on 3D convolution neural network (3D-CNN). In the sequential model based on the deep learning framework of Keras, 3D convolution, pooling, batch normalization and other layers are added to construct the sequence. The recognition rate of this model on SMIC database can reach 76.92%, and it also shows good recognition rate on other databases. This method is superior to or partially superior to the classical methods and the current mainstream methods.