Semi-Dynamic Facial Expression Recognition based on Masked Displacement Image

Facial expression recognition, as an interesting problem in pattern recognition and computer vision, has been performed by means of dynamic and static methods in recent years. Though dynamic information plays important role in facial expression recognition, utilizing the entire dynamic information of expression image sequences have high computational cost compared to the static cases. In order to reduce the computational cost, only neutral and emotional faces can be used instead of entire image sequence. In the previous research, this idea has been employed by means of DLBPHS method which vanish facial important small displacements by subtracting LBP features of neutral and emotional face images. In this paper, a novel semi-dynamic approach is proposed to utilize two face images by subtracting neutral image from emotional image. In this method, LBP features are extracted from masked difference image. In the masking procedure, eyes and mouth being important regions in facial expression recognition are selected for feature extraction step. Evaluation of the proposed algorithm on the standard databases shows a significant accuracy improvement compared to DLBPHS method.