Research On Micro-expression Recognition Algorithm Based On Apex Feature

Facial expressions are an important way for people to express their inner activities and state.According to different facial activities,facial expressions can be divided into macro expressions and micro expressions.Macro facial expressions are facial expressions that people can directly detect,but to some extent,they are deceptive.Micro-expressions have spontaneous,transient,and subtle characteristics,which can better reveal potential inner emotions and play an important role in analyzing and studying human emotions.Therefore,they are considered one of the important clues for detecting inner activities and have wide applications in fields such as emotional states,lie detection,and business negotiations.Due to its uncontrollability and small amplitude,the recognition of microexpressions poses significant challenges.How to accurately recognize micro-expressions is currently a hot topic of research.In this paper,analyzes the characteristics of micro-expressions,summarizes the characteristics of micro-expressions,and uses Apex features and deep learning methods for micro-expression recognition.The Apex frame is the apex frame in the micro-expression sequence,which contains a lot of key micro-expression information.A model suitable for micro expression recognition was proposed using a dataset of selfgenerated micro expressions under laboratory control as the research object to address the problem of difficulty in recognition and low recognition accuracy due to the small number of samples and uneven distribution of different categories in micro expression recognition.The application of Apex feature based recognition methods in micro expression recognition tasks was studied.The main content is as follows:(1)Using apex frames and improved residual networks for microexpression recognition.Extracting Apex apex frames containing more critical information from micro expression video sequences;The residual network improved by adding SE module is used for feature extraction of apex frames of micro-expressions.The SE module can better learn key information in the features,and Res Ne Xt replaces dense structures with sparse structures through group convolution to simplify the structure,reduce the number of parameters,and improve recognition efficiency.Simulation experiments were conducted on the micro expression dataset,and various micro expression recognition methods were compared based on the same evaluation criteria.It was found that the improved residual network and apex frame improved the accuracy and F1 value of micro expression recognition.Further ablation experiments showed that using the improved residual network and apex frame simultaneously also improved accuracy and F1 value compared to not using the method.The experiment shows that the improved residual network and apex frames can reduce the impact of a small dataset,and the model has good fitting performance,which can improve the recognition accuracy of micro expressions.(2)Use the optical flow features of apex frames for micro expression recognition.Add the optical flow information of the apex frame to the proposed recognition model,and perform feature extraction and recognition operations together with the apex frame.The extracted apex frame information includes vertical and horizontal optical flow features from the start frame to the apex frame,and vertical and horizontal optical flow features from the apex frame to the end frame;Connect the extracted apex frame optical flow features with the apex frame as input,and input them into the improved residual network added to the SE module for operation.Increasing the optical flow information of apex frames can further enhance the temporal features of micro-expressions.According to the experimental results,adding richer features can further improve the accuracy of recognition compared to F1.At the same time,the effectiveness of optical flow features and some structures in the network was demonstrated through ablation experiments,verifying that the proposed method can further improve the fitting effect of the model,improve the problem of significant performance differences in different categories,and have better recognition performance,enhancing the application and research value of micro-expression recognition..