基于TBANet的人脸微表情识别方法

打开文本图片集

中图分类号：TP391.4 文献标识码：A 文章编号：2096-4706（2025）16-0034-05

Abstract： Micro-expressionrecognitioncanreveal thereal emotional state individuals，so it shows a widerange applicationprospectsinthefeldsumanomputerinteraction，psholgicaldiagnsisinalinvetigation.Duetoth smallampliudeadsortduraionmicro-expressions，thetradiionalneuraletork modelisdiffculttoefeivelycapture extractthekeyfeaturesmicro-expressons，thuslimitingtheimprovementrecognitionaccuracyInordertosolvethese problems，this paper proposesanew neuralnetworkarchitecture——TBANet （Transformer Block Aggregation Network）. Firstly，thehorzontalverticalopticalfowimagesmicroexpresionsaecalculatedbyRAFTalgorithmtocharacterize facial motion information.Secondly，the hierarchical Transformer network isused to encode the motion relationship each partthefacewenthemicro-expressionoccurs，ndthefeaturerepresentationabilityisfurtherehanced.FinallyheBlock Aggregation layer isusedtusethe hierarchicalstructure features toachieve multi-sale informationcapture.Experimntal resultsonthreedifferentdatasetsverifythe effectivenessTBANet.

Keywords： micro-expresson recognition; Transformer; RAFT; optical flow; feature fusion

0 引言

面部表情在辨别人类情感方面起着至关重要的作用[1]。（剩余9214字）