Research On Micro-expression Recognition Method Based On Deep Forest

Micro-expression is a kind of fast and faint facial movement.It is a short-term,involuntary expression on human's face when trying to cover up or suppress a certain emotion.When people realize that facial expressions are happening,they may try to suppress facial expressions because the emotions they show may be inappropriate.Once suppression occurs,people will cover up the original facial expressions and cause micro-expressions.People's hidden emotions and their intentions can be revealed by micro-expressions,so micro-expressions can potentially be used in applications such as medical diagnosis,road safety,criminal interrogation,polygraph detection and business negotiations.There are certain difficulties in the recognition of microexpressions,caused by the short duration of micro-expressions and the low intensity of facial movements.The current research methods of micro-expression recognition are mainly concentrated in the field of traditional algorithms and deep learning.On the issue of microexpression recognition,although people have done a lot of researches in the field of traditional algorithms and deep learning,it is difficult to obtain a recognition rate as high as that of macroexpression,which is still far from the actual application.The micro-expression recognition rate is not high,it has the problem of a small data set and contains redundant information.Based on the existing method,a micro-expression recognition method based on peak frame features is proposed.The method includes four parts:the first part is the micro-expression sequence pre-processing model,the second part is the peak frame selection model,the third part is the deep feature extraction model,and the last part is the deep forest micro-expression classification model.The method selects the peak frame,the sub-peak frame,and the middle frame of the two frames as the training set,thus it can obtain a larger data set than using only the peak frame as the training set.The method can not lead to the introduction of too many general micro-expression frames and information redundancy.At the same time,considering the respective advantages of deep convolutional neural networks and deep forests,this paper proposes a micro-expression recognition training model.This model combines deep convolutional neural networks and deep forests.It automatically extracts features through deep convolutional neural networks based on transfer learning and combines deep forests in small samples.It brings an excellent performance of classification prediction on the data set,thereby improving the recognition rate of micro-expressions.Finally,for the purpose of verifying the effectiveness of the proposed method,the average absolute error is used as the peak frame selection evaluation criterion.The recognition rate,precision rate,recall rate,and F1-measure are used as the evaluation criteria for microexpression recognition,and simultaneously the confusion matrix is used to represent micro confusion between categories in expression classification results.In this paper,a simple deep forest model composed of different cascade classifiers is compared under different super parameters.The micro-expression recognition results of the deep forest model trained using three frames above are compared with the recognition results of the model using only the peak frame or the entire frame.The experimental results are also compared with the experimental results of the existing methods.The results show that the micro-expression recognition method based on deep forest proposed in this paper can improve the micro-expression recognition rate to a certain extent and is feasible.