Facial Expression Recognition Based on Local Vector Model

Texture feature extraction is an important step in the facial expression recognition system. The traditional LBP method ignored the statistical characteristics of the texture change direction in the process of feature extraction, and we can extract more detailed texture information by the LDP method based on LBP, but the computational complexity is greatly increased. In order to extract more detailed texture information with the computational complexity is not increased, we proposed a method named Local Vector Model (LVM). In this method, modulus value and direction of the local texture changes are extracted as the features of classification. Furthermore, in order to improve the robustness that the algorithm to the subtle deformation of expression image, the Image Euclidean Distance is introduced and embedded in LVM. Finally, the even decreasing function is used to get the neighbor classification distance. Experiments on JAFFE facial expression databases with different resolution demonstrated that the method proposed in this paper is better than other modern methods.