DynGeoNet

Micro-expressions (MEs) are brief and involuntary facial expressions when people hide their true feelings or conceal their emotions. Based on psychology research, MEs play an important role in understanding genuine emotions, which leads to many potential applications. However, the ME analysis system can still not work well in the real environment because of the challenging performance of ME spotting, which is to spot the images with micro-expressions from long video sequences. To improve the performance of ME spotting, we focus on hybrid feature engineering, which aims to create a robust feature for discriminating tiny movements. The proposed framework consists of two main modules: (1) the feature engineering extracts both geometric features and appearance features based on dynamic image; (2) the new deep neural network inputs the handcrafted feature for the late fusion and ME samples classification. Our experimental results from three baseline datasets demonstrate the promising results.