Facial Micro-Expression Recognition in Video using Squeezed Landmark Feature Maps

Micro-Expression is the muscle movements of face that appear unconsciously. It is difficult to observe the Micro-Expression because of its short duration and low intensity. In recent work, a two-dimensional (2D) Landmark Feature Map (LFM) that is able to represent spatiotemporal feature of Micro-Expression was proposed. The LFM handles the duplicate values since there are 68 Landmark points in LFM and each point measures the distance from all remaining points. These values raise the problem of increasing processing time, cost complexity, and CNN network size. we propose a Squeezed LFM and a Squeezed CLFM that solves the problems. We also evaluate the performance of the proposed method using a public micro-expression dataset SMIC. The proposed method achieves the reduction of the data size and CNN network parameter amount without any recognition accuracy degradation of Micro-Expression.