| Micro-expressions have been used as deception detection cues in many applications,such as customs security and clinical diagnosis.However,there is still no definite conclusion on whether they can reliably and effectively detect deception.The study of the relationship between micro-expressions and deception needs to be supported by a large amount of data.However,the existing micro-expression database is still inadequate.Therefore,this study aims to construct a high ecological validity micro-expression database in deception contexts so that scholars can subsequently conduct micro-expression and deception detection studies.Study 1 captured deception and honesty video clips of 124 college students(Mean age=21.41,SD=3.18)through the deception game paradigm.Then used Premiere Pro to edit the video and process the deception and honest video clips by ELAN for manual annotation,and rapid facial movements of less than 500 ms duration were labeled as a micro-expression sample,and the AU(s)of each micro-expression were labeled according to the Facial Action Coding System(FACS).The AU occurrence frames were also calculated using Openface and MATLAB software,and the AU motion of consecutive frames<10 was counted as one micro-expression within 200 ms,and the AU motion of consecutive frames<25 was counted as one micro-expression within 500 ms.i.e.,the number of micro-expressions in each subject’s honest and deceitful videos was calculated by the computer method.Finally,descriptive statistics and paired-samples t-tests were conducted on the number of micro-expressions leaked in deception and honesty contexts.And the results of the manual labeling were compared with the results of the computer analysis.Study 2 used material from the published deception database C2W2 D for micro-expression coding annotation and computer analysis,which consisted of data from 26 drug addicts(Mean age=36.21,SD=8.43)and 58 healthy college students(Mean age=20.54,SD=2.25),and was processed and analyzed in the same way as Study 1.Results:1.The paired-sample t-test for the average number of micro-expressions per 1000 frames in manually labeled deceptive,honest situations in the deception game paradigm showed significant differences(t=-4.275,p=0.001),and the number of micro-expressions appearing in the deception context is 0.5 more per 1000 frames than in the honest context.2.The results of the paired-sample t-test for the number of micro-expressions within 200 ms per100s on average in the deception,honesty context derived from computer analysis under the deception game paradigm showed that AU07(t=2.283,p=0.024),AU25(t=4.565,p=0.000),AU26(t=2.518,p=0.013),AU28(t=2.010,p=0.047)indicators were significantly different,while the remaining indicators were not significantly different.t-test results for the number of paired samples of micro-expressions within 500 ms showed that AU07(t=2.563,p=0.012),AU14(t=2.290,p=0.024),AU25(t=4.697,p=0.000)AU26(t=2.661,p=0.009)indicators were significantly different,and the rest of the indicators were not significantly different.More microexpressions appeared in the deception context than in the honesty context.3.The results of the paired-sample t-test for the mean number of micro-expressions within200 ms per 100 s of the drug addicts in the deception and honesty contexts derived from computer analysis under the werewolf game paradigm showed significant differences in the AU17(t=2.835,p=0.09)indicator and no significant differences in the remaining indicators.The results of paired-sample t-test for the mean number of micro-expressions within 500 ms per 100 s of the detoxified persons showed that there was a significant difference in the indicators AU07(t=2.196,p=0.038),AU17(t=3.737,p=0.001),AU25(t=4.228,p=0.000),and no significant difference in the rest of the indicators.Less microexpressions appeared in the deception context than in the honesty context.4.The results of the paired-sample t-test for the average number of micro-expressions within200 ms per 100 s for college student subjects in the deception and honesty contexts derived from computer analysis under the werewolf game paradigm showed significant differences in the indicators AU23(t=2.876,p=0.006),AU25(t=3.659,p=0.001),and no significant differences in the remaining indicators.The results of paired-sample t-test for the average number of micro-expressions within 500 ms per 100 s for college student subjects showed significant differences for the indicators AU23(t=2.705,p=0.009),AU25(t=4.399,p=0.000),and no significant differences for the remaining indicators.Less microexpressions appeared in the deception context than in the honesty context.Conclusion:In this study,a database of spontaneous micro-expressions in deception situations was constructed,including a total of 201 micro-expression samples(78 honest and 123 deceptive)from 16 college students and 8 micro-expression samples(4 honest and 4 deceptive)from 4 drug addicts,divided into two parts A and B.Part A is the display of micro-expressions in long videos,which can be used for automatic micro-expression recognition by computer vision algorithms to for deception detection;Part B is separate micro-expression samples,which can be used for micro-expression recognition training.Also,analysis of the database leads to the conclusion that individuals produce micro-expressions in both honest and deceptive contexts.In some cases more micro-expressions were produced in the deception context than in the honesty context,but the results differed depending on the type of data. |
PDF Full Text Request