Automatic 3D facial expression recognition based on polytypic Local Binary Pattern

Most of the existing methods adopted in 3D facial expression recognition require manual-intervention, which should be avoided in automatic systems. In this paper, we propose an automatic data normalization method based on the point-to-point mapping between the depth value and the texture information of the facial data. Based on this, we further construct a new kind of 3D facial expression feature, namely polytypic Local Binary Patterns (p-LBP), which involves both the irregular divisions to keep the integrity of local faces and the fusion of depth and texture information of 3D models to enhance the recognition accuracy. The proposed strategy is tested on BU-3DFE database, and three kinds of classifiers are employed. Their remarkable results outperform the state of the art, and show the effectiveness of proposed features for 3D facial expression recognition. Therefore, p-LBP is validated and its simplicity opens a promising direction for automatic 3D facial expression recognition.