Micro-Expression Recognition Based on Local Two-Order Gradient Pattern

Recently, micro-expression recognition has obtained more and more attention due to its spontaneity. Moreover, how to extract feature from these subtle image sequences is still a hard issue. Currently, a lot of local patterns have been applied to describe these micro-changes, but in essence they are seldom specialized to capture fine structure changes in video clips and often need tens of frames to generate a descriptor through interpolation between frames. To make up for these deficiencies, this paper proposes a novel local pattern i.e. Local Two-Order Gradient Pattern (LTOGP) to describe the subtle changes in micro-expression sequences. Using this pattern, we can not only capture subtle changes in videos, but also only make use of 5 frames to generate a descriptor of a micro-expression clip rather than tens or hundreds of frames in related literatures. Experiments are conducted on two databases of CASME and CASME2 and the results show that our proposed method gains promising performance for micro-expression recognition with proposed method.