Micro-Expressions Detection Based on Micro-Motions Dense Optical Flows

Micro-expressions are one of the most reliable sources of deceit detection. They are fast, involuntary facial movements and they usually occur when a person tries to hide their true feelings. This paper presents a novel method for micro-expression detection in high speed video-sequences. We first define several free-form regions of interest on the face based on the location of the muscles of facial expression. Dense optical flow features are extracted from each region, and their magnitude and direction are used to spot micro-expressions in the video sequence. The direction of movement is used to filter out irrelevant, macro movements. The proposed algorithm for micro-expression detection is training free and it involves a simple voting procedure on the movement magnitude from each region. The proposed method was tested on a publicly available database.