Study of DRMF and ASM facial landmark point for micro expression recognition using KLT tracking point feature

Abstract

Micro-expression recognition is one of the popular researches in analysing expressions on the face. Micro-expression is a facial movement that occurs in a short time and is difficult to identify manually by human eyes. In general research, facial landmarks are used to form a large size ROI for each facial feature for the feature extraction process. In this study, we track the subtle motions of micro expressions by using point features. This approach aims to get feature extraction from tracking results and then analyse micro-expression. We compared the Active Shape Model and Response Map Fitting methods to produce accurate points and fast time on facial features. To measure the subtle motion tracking of facial features in each frame tracking is done using the Kanade-Lucas-Tomasi method. To estimate the rationality of our method, we conducted an experiment on CASME II and SAMM dataset for micro-expressions. The results show that the points on DRMF are more accurate with point-to-point error is 7.9 and the time taken is faster which requires time is 0.02 second. We evaluated the method proposed for evaluation showed that using CASME II - Naive Bayes (79.3%) and SAMM - Naive Bayes (74.6%).