The new 3D facial expression recognition method based on semantic knowledge of Gaussian mixture model

Firstly, 2D images is susceptible to the face-pose, illumination etc. Secondly, image recognition are mostly based on image low-level visual features, while human perception of images are based on high-level semantic knowledge, which results in the "semantic gap" between them. For this reason, a new 3D facial expression recognition method is proposed based on semantic knowledge of Gaussian mixture model. The method uses Gaussian curvature and mean curvature to extract several key points of low-level visual features of 3D facial expressions, and uses European-style distance to form several key points into a set of low-level visual feature vectors. Then the Gaussian mixture model and the AHP hierarchical model are combined to calculate the high-level semantic feature vector, which solves the "semantic gap" between the low-level visual features and the high-level semantic knowledge of facial expression images, and improve the robustness and recognition rate of 3D facial expression recognition.