Facial Micro-Expression Spotting From Videos

Beh, Kai Xin (2019) Facial Micro-Expression Spotting From Videos. Final Year Project (Bachelor), Tunku Abdul Rahman University College.

Text Beh Kai Xin.pdf Restricted to Registered users only Download (1MB)

Abstract

Micro-expressions are involuntary facial expressions that last between 1/25th to 1/5th of a second and it usually happens when people try to mask their emotions in high-stake situations. As a result, it is hard to detect the occurrence of a spontaneous micro-expression because the limitation of human vision in spotting the brief and subtle change of facial expression. Nonetheless, the study of the micro-expressions has gaining more attentions lately because there are a number of impressive potential contribution can be derived from the application of micro-expressions recognition despite the difficulty in recognizing the micro-expression. For instance, individual who works in law enforcement can detect lies told by a criminal by reading the criminal’s micro-expression, health professional can produce a more accurate diagnosis of their mentally-ill patient by obtaining complete information through recognizing their patient’s micro-expression, and many more. In order to cope with the limitation of human vision in reading the micro-expressions, a computer-based micro-expression spotting system is suggested in this proposal. A micro-expression recognition system can be categorized into two major tasks which are the spotting of the fleeting change of facial expression and the classification of the emotion behind the spotted micro-expressions. In this thesis, the proposed system is solely focus on the spotting part because spotting of the micro-expression is the primitive procedure in the micro-expression recognition system. Last but not least, the detection of the change in the ratio of the Euclidean distance computed at the selected facial regions such as eyebrow and mouth between video frames that last for 8 consecutive frames will be indicated the presence of the micro-expressions. In a nutshell, the average accuracy rate and the highest accuracy rate achieved by this system are 64.77 % and 82.30 % respectively.