| Nowadays face recognition, and facial expression recognition has become a very popular area in computer vision research, which has achieved significant progress in the research field. In this work, we present a new approach local pattern descriptor, Local Gradient Derivative Pattern (LGDP), to face recognition which considers more detailed information than the Local Binary Pattern (LBP) and has a better code rule compare with LBP. The proposed method is focus on how to solve the problem of illumination and pose rotated. In order to solve these two problems of the face recognition, the main direction and new code rule are pulled into our research. We get the gradient image from the original image and then we calculate the main direction based on this gradient image then to code the binary always according to the new code rule. In this method the face image is first divided into several small regions from which Local Gradient Derivative Pattern (LGDP) histograms are extracted and concatenated into a single, spatially enhanced feature vector to be used as a face descriptor. In this method we use weighted Chi square distance for matching, in this method we only set to value one is one and the other is two, value one means this area is not very importance but the value two is opposite. The face recognition is performed using a nearest neighbor classifier in the computed feature space with Chi square as a dissimilarity measure. To demonstrate the robust of this new method, I have done some experiments. Three well-known and challenge- ORL, Yale and FERET face databases are used in the performances to evaluate the method. And in the experiment we also use the Leave-one-Out method to test our method, this experiment method can always be seen in the paper about the face recognition. The experiment compares the proposed algorithm with some traditional methods. The experiments result clearly show that the proposed method give us a better performance than some other methods. |
PDF Full Text Request