A main directional maximal difference analysis for spotting facial movements from long-term videos

摘要

There is an increasing interests in micro-expression researches. Spotting micro-expressions in long-term videos is very important, not only for providing clues for lie detection, but also for reducing the labor required to collect micro-expression data. However, little progress has been made in spotting micro-expressions. In this paper, we propose a Main Directional Maximal Difference (MDMD) Analysis for micro-expression spotting. MDMD uses the magnitude maximal difference in the main direction of optical flow features to spot facial movements, including micro-expressions. Using block structured facial regions, MDMD obtains more accurate features of movement of expressions for automatically spotting micro-expressions and macro-expressions from videos. This method involves both the temporal and spatial locations of face movements. Evaluations using the CAS(ME)(2) database containing micro-expressions and macro-expressions show that MDMD is more robust than some state-of-the-art algorithms.