Research On The Theory And Method Of Micro-Expression Recognition

Micro-expression is a kind of expression of very short duration, which can express the true feelings that a person wants to hide. In recent years, micro-expression recognition has been widely concerned because of its potential to apply. At present, the related research is at its early stage due to the following two reasons. Firstly, there is shortage in micro-expression database. Secondly, most feature extraction methods are not robust because of the nuances in micro-expression. According to these problems, with the inspiration of latest deep learning success, this article has explored and studied the method and application of micro-expression recognition research. The main work is summarized as follows:(1) Established an induced micro-expression database. Micro-expression database is important support for micro expression recognition research, and currently induced micro-expression database is too scarce.This paper set up the induced environment for micro-expression, used video cameras to collect micro-expression video of testees’ watching induced videos, picked the micro-expression frame sequence, marked the micro-expression, set up micro-expression database, and summarized some of the challenges for building a database.(2) Summarized a set of complete micro-expression recognition work, including micro-expression frame sequence pretreatment, micro-expression detection, micro-expression feature extraction, and classification of micro-expression. A series of benchmark experiments were conducted.(3) Proposed a micro-expression recognition method based on Deep Belief Network.Deep learning and micro-expression recognition study were combined together. Sample data of the micro-expression was extended. Then the dynamic characteristics were extracted into Deep Belief Network. Parameters were adjusted in the process of the preliminary training and fine-tuning and finally a better recognition rate has been achieved.(4) Proposed a micro-expression recognition method based on three-dimensional Convolutional Neural Network. At present, Convolutional Neural Network is one of the most commonly used deep learning Network model for studying pattern recognition problems. However, CNN’s ability is only limited in the processing of 2D input. In this paper, on the basis of CNN, a three-dimensional Convolutional Neural Network has been proposed to build the Network structure, and is used to extract the feature of dynamic videos of micro-expression. Classification experiments based on these features are also conducted. The innovation of 3D-CNN lies in the fact that it is a new kind of deep learning network structure which increases the convolution of time information. It can deal with all kinds of 3D input, extract dynamic features and also can be used for dynamic identification directly.(5) An automatic detection and recognition system for micro-expression has been set up. The system is divided into two parts. The micro-expression detection system can detect the starting frame, peak frame, end frame of micro-expression video, and micro-expression identification can classify five kinds of emotions from micro-expression videos.