Facial Expression Recognition Based On The Fusion Of ASM Differential Texture Features And LDP Features

Facial expression recognition is an important part of theory research of artificial psychology and emotion, involves in image processing and analysis, pattern recognition, computer vision, computer graphics, artificial intelligence, cognitive science, biology and other sciences. The deep research of facial expression recognition can promote the development and progress of science. Facial expressions are divided by psychologist into six basic categories:anger, disgust, fear, happy, sad and surprise. In this paper, we conduct the experiments of feature extraction and classification of expression on six basic expressions. Because of the complexity of expression, it is difficult to describe and express the expressions’ using single features. So in our paper, we mainly study the two kinds of expression features extract method, and combine the two kinds of features in decision level to improve the expression recognition problem. Our main work is as follows:1. A novel method based on DT (namely, Differential Texture) features for facial expression recognition is proposed. First, we establish a standard reference model, in which there are 55 landmark points. These feature points are mainly located at the mouth, eyes, nose, eyebrows and facial contour. Then we use respectively ASM (Active Shape Model) to get the positions of 55 landmark points for neutral expression and non-neutral expression. After that, the Delaunay Triangulation is applied to these feature points, and the textures in expression image are warped to the standard reference model. Finally, we obtain the difference of neutral expression and non-neutral expression after above processing as the DT features. The experimental results show that we get a good recognition rate by using DT features for facial expression recognition.2. We study the facial expression recognition method based on LDP (Local Directional Pattern) features and introduce its varieties:LDPv and LSDP. An LDP feature obtained by computing the edges response values in 8 directions at each pixel and encoding them into an 8 bit binary number, so it is a robust feature descriptor. It is easy to extract the edge information. The experimental results show that we can obtain a high recognition rate and a good robustness by using the LDP.3. We propose a new method for facial expression recognition, which combines DT features and LDP features in decision-making level by DS (Dempster-Shafer) evidence theory. DS evidence theory can deal with uncertainty problems. In the DS evidence theory, the uncertain probability can be directly allocated to unknown even, that is to say, it can be assigned to any proposition in recognition framework (?). The experimental results show that this method can improve the robustness and recognition rate of facial expression to some extent.