Micro-Facial Movement Detection Using LBP-TOP Descriptors for Landmark Based Regions

In the paper the problem of the micro-facial movement detection is considered. The modern approaches to different stages of the micro-expression recognition are considered. A micro-facial movement detection pipeline based on the calculation of the LBP-TOP descriptors for the local 3D blocks is proposed. Two ways of constructing these blocks around the facial landmarks in the regions of the possible Action Units (AUs) is also proposed. The selection of the methods for each stage of the proposed pipeline is explained. The experiments carried out on the Spontaneous Actions and Micro-Movements (SAMM) dataset are described. The performance of the micro-facial movement detection from 74% to 81% is achieved depending on the type of the used classifier and the 3D block construction algorithm in the proposed pipeline.