Micro-expression recognition using local binary pattern from five intersecting planes

Micro-expression recognition has important research value and huge research difficulties. Local Binary Pattern from Three Orthogonal Planes (LBP-TOP) is a common and effective feature in micro-expression recognition. However, LBP-TOP only extracts the dynamic texture features in the horizontal and vertical directions and does not consider muscle movement in the oblique direction. In this paper, the feature in oblique directions is studied, and a new feature called Local Binary Pattern from Five Intersecting Planes (LBP-FIP) is proposed by analyzing the movement direction of facial muscles in the micro-expression video. LBP-FIP concatenates the proposed Eight Vertices LBP (EVLBP) with LBP-TOP extracted from three planes, where EVLBP is extracted from two planes in the oblique direction. In this way, the dynamic texture features in the oblique direction are extracted more directly. On the CASME II and SMIC database, we evaluated the proposed feature and the effectiveness of the features in the oblique direction. Extensive experiments prove that LBP-FIP provides more effective feature information than LBP-TOP, and extracting the features in oblique directions is discriminative for recognizing micro-expressions. Also, LBP-FIP has advantages comparing with other LBP based features and achieves satisfactory performance, especially on CASME II.