Facial Micro Expression Recognition for Feature Point Tracking using Apex Frames on CASME II Database

Abstract

Micro-expressions are facial expressions that occur inadvertently to hide true feelings (emotional leaks). Although previous studies used the entire face area and all frames in the video dataset, this resulted in relatively long computation time and data redundancy. The main contribution of this research is to apply recognition micro-expression analysis using a comparison of apex frames with manual (handcrafted) and random sampling of frames and applying feature point tracking to the brow area and corners of the lips. The method for forming feature points in the facial area uses Discriminative Response Map Fitting (DRMF), then facial feature points are tracked using Kanade-Lucas-Tomasi (KLT). This feature point tracking produces motion feature data as feature extraction data. Finally, a comparative analysis of the classification method using the Support Vector Machine (SVM) and MLP-Backpropagation was conducted using the CASME II dataset. The experimental results of this study show significant results with an accuracy of 81.3% on MLP-Backpropagation and an average computing time of 1.45 seconds for each video. From the results of this study, information on the apex phase can contribute information that is very important for facial micro-expression recognition.