Abstract
Facial micro-expressions are brief, involuntary expressions appeared on the face of people who are trying to conceal or repress emotions. This paper developed a questionnaire containing control questions and relevant questions. We analyze the motions of key features on the faces and the appearance of the micro-expressions, we quantify the characteristic of telling lies by observing a sequence of frames extracted from videos. In the model, we test several factors which including eyes’ and mouth’s micro movements, and then fit the logistic regression model to find the main effect on the facial movements when telling lies. The empirical results indicate that people tend to lie while decreasing the horizontal length of their mouths. Furthermore, the vertical width of eyes tends to change significantly in a very short period while lying. The classification result of model shows that sensitivity = 74.36% as the proportion of deceptive response being classified correctly; and the overall classification accuracy is 62.23%. We also find out that when people are answering high-stakes questions, our model will achieve a 71.79% classification accuracy in deception response with an overall accuracy as 65.38%, which is higher than the accuracy of all data.