Micro-expression recognition based on the fusion between optical flow and dynamic image

Abstract

Micro-expression (ME) is a subtle and involuntary facial expression that reveals the human's concealed emotion. Psychology studies pointed out that research of ME can build potential applications in many fields. Therefore, ME recognition (MER), one of the two main ME analysis tasks, has been becoming an attractive topic recently. However, the work of MER is still needed to consider due to several limitations related to performance and dataset. This paper proposes a feature fusion between optical flow and dynamic image to create a robust ME representation, which can be learned effectively from deep learning techniques. Experiments from two public datasets, CASME-II and SAMM, show that our method obtains higher performance than several existing studies and is very promising for future research. CCS CONCEPTS •Computing methodologies∼Object recognition