面向视频的人脸特征计算方法.pdfVIP

431Vol.43No.1

20251JOURNALOFAPPLIEDSCIENCES—ElectronicsandInformationEngineeringJan.2025

DOI:10.3969/j.issn.0255-8297.2025.01.010

102600

TP390255-8297(2025)01-0137-17

Video-BasedFacialFeatureComputationMethods

WANGYingxiao,YANGYanhong,TANYunfeng

SchoolofInformationEngineering,BeijingInstituteofGraphicCommunication,

Beijing102600,China

Abstract:Thispaperpresentsareviewofresearchinvideofacerecognitionconductedover

thepastfiveyears.Itprovidesacomparativeanalysisofthefacialfeaturecomputation

methods,categorizingthemintotraditionalapproaches,deeplearningtechniques,and

featureaggregation/fusionmethods.Traditionalfeatureextractionmethodsincludelinear

andnonlinearapproaches,whiledeeplearningmethodsincludespatialandtemporalfeature

extractiontechniques.Featureaggregationandfusionmethodsintegratemultiplefeature

sourcesandfusefeaturesfromdifferenttimeperiodstoimproverecognitionperformance.

Attheendofeachsubsection,thispaperalsoprovidesaunifiedanalysisofthealgorithms

usedintheliterature,highlightingtheiradvantages,evaluationmetrics,andapplications.

Throughthisresearch,weaimtoprovidemorereliableandefficientsolutionsforpractical

applicationsofvideo