Research On Establishment Of Micro-Expression Database And Technology Of Micro-Expression Detection

Micro-expression,distinguishing from macro-expression for its brief duration,weak motion range and less action areas,is a facial expression out of psychological control.Currently,the tremendous potential of micro-expressions at the area of national security,judicial trail and lie test has aroused great concern and the related researches have some achievements aid by the development of computer graphics technologies.However,the sample insufficiency and quality irregularity of the cornerstone micro-expression database caused by the difficulties in expression capturing,the complication in collecting,the waste of time and labor consumption in encoding and the lack of evaluation criteria have become a restriction for intensive studies on the detection and recognition of micro-expressions.To address the above problems,in this paper,a micro-expression database SDU with largest sample quantities and highest resolution is established to overcome the shortcomings such as sample insufficiency,low resolution and optical unnaturalness,meanwhile,an assessment method of the database quality is proposed by the vertical comparison of the database itself and the horizontal comparisons with other databases.Finally,based on the deep analysis of the features of SDU micro-expression database,two micro-expression automatic detection methods are proposed:the key points extraction one based on the joint cascade method and the interested regions extraction one based on the feature amplitudes and angles of optical features.The main work and innovation points in this paper include the following aspects:Firstly,SDU micro-expressions database is established including 300 samples of 7 emotions with high qualities.It is obtained by elaborate selection of stimuli,shooting environment and camera equipment,the participation of psychology students in sample analysis and the full integration of psychological knowledge under the guidance of psychological doctors.This database,which possesses comprehensive classification,excellent quality and largest quantity,is a favorable experimental material for detection and recognition of micro-expressions.Secondly,an assessment method for micro-expression databases is proposed aiming at the standard deficiency and sample irregularity status.Three evaluation perspectives,the subjective evaluation,objective evaluation and the image feature analysis,are adopted in this paper,and from each perspective,the quality of SDU database is evaluated specifically from different aspects such as resolution,frame rate,and coding bit rate.At last,according to the evaluation conclusion,the qualities of SDU database are evaluated.Thirdly,to solve the problem of high labor and time consumption of artificial encoding in the process of establishment,two automatic detection methods from video sequences are proposed.One is to contrast the deformation patterns of interested feature points by setting proper threshold based on the joint cascade ideology;the other is to compare the change pattern of the features of interested areas by setting different threshold for different parts of human face based on the amplitude and angle information extracted by optical flow method.The two micro-expression detection methods proposed in this paper all obtained fair detection results through the experimental analysis on the SDU micro-expression database and the current most used CASME? micro-expression database.