Adaptive Mask for Region-based Facial Micro-Expression Recognition

Abstract

Facial micro-expression can be characterized by its short duration and subtle movements. In facial micro-expression recognition, these subtle movements require more specific feature descriptors due to only a few parts of the face produce information that helps us to recognize micro-expressions. Over the past decade, researchers designed different Region of Interests (ROIs) to study specific face regions in micro-expressions recognition. To further study this aspect, we proposed a region-based method with an adaptive mask for facial micro-expression recognition. Based on the most frequent Action Units on the two publicly available datasets, i.e. CASME II and SAMM, 14 ROIs are defined where the adaptive mask is created by calculating the optical flow after Gaussian smoothing. Further, LBP-TOP features are extracted from each ROIs and Sequential Minimal Optimization is used to classify the micro-expressions. When evaluating our proposed method on CASME II and SAMM, we achieved the accuracy of 68.2% and 56.1%. In terms of F1-Score, our proposed method achieved 0.57 on CASMEII and the best performance of 0.50 on SAMM.