Spatiotemporal local orientational binary patterns for facial expression recognition from video sequences

On account of encoding the binary result of the differential values between the central pixel and its neighbors, the local binary pattern (LBP) cannot describe more detailed information. Aiming at this issue, local orientational binary patterns (LOBP) is proposed. LOBP encodes the variations of the neighborhood orientational difference, which represents more detailed pattern information in the local region. An extension of LOBP to 3-D space is proposed, which is called spatiotemporal local orientational binary patterns (SLOBP). SLOBP uses three orthogonal planes (XY, XT, and YT) that intersect in the center pixel. SLOBP considers the feature distributions obtained separately from each plane and then concatenates them together. SLOBP can capture effectively the spatiotemporal feature in 3-D space such as video sequence. Experimental results demonstrate that the proposed method outperforms other modern approach applied in facial expression recognition from video sequences.