Facial recognition using new LBP representations

In this paper, we propose a facial recognition based on the LBP operator. We divide the face into non-overlapped regions. After that, we classify a training set using each region at a time under different configurations of the LBP operator. Regarding to the best recognition rate, we consider a weight and specific LBP configuration to the regions. To represent the face image, we extract LBP histograms with the specific configuration (radius and neighbors) and concatenate them into feature histogram. We propose a multi-resolution approach, to gather local and global information and improve the recognition rate. To evaluate our proposed approach, we considered the FERET data set, which includes different facial expressions, lighting, and aging of the subjects. In addition, weighted Chi-2 is considered as a dissimilarity measure. The experimental results show a considerable improvement against the original idea.