Abstract

Micro-expressions are rapid and subtle facial movements such that they are difficult to detect and recognize. Most of recent works have attempted to recognize micro-expression by using the spatial and dynamic information from the video clip. Physiological studies have demonstrated that the apex frame can convey the most emotion expressed in facial expression. It may be reasonable to use apex frame for improving micro-expression recognition. However, it is wonder how much apex frames contribute to micro-expression recognition. In this paper, we primarily focus on resolving the contribution-level by using apex frame for micro-expression recognition. Firstly, we propose a new method to detect the apex frame in frequency domain, as it is found that apex frame has very correlated relationship with the amplitude change in frequency domain. Secondly, we propose to use deep convolutional neural network (DCNN) on apex frame to recognize micro-expression. Intensive experimental results on CASME II database shows that our method has achieved considerably improvement compared with the state-of-the-art methods in micro-expression recognition. These results also demonstrate that apex frame can express the major emotion in micro-expression.