Wearing-Glasses Face Recognition Based On Local Gabor Binary Pattern

Face recognition is one of the hottest research fields in pattern recognition. Recently, with the development of high-speed hardware and requirements of business or security, research on face recognition achieved a great progress. However, under uncontrolled environments, which contain illumination, expression various or occlusion (like eyeglasses), recognition results are not very good. In reality, it is very common for face images including eyeglasses. Therefore, it is very important for dealing with wear-eyeglasses problem. To improve the performance of face recognition, this paper will focus on the glasses-face recognition.So far, methods dealing with glasses problem just have removed the glasses or regarded glasses as general occlusion according to using local feature-extracted approaches for recognition. However, these methods appeared respective shortcomings. To improve the recognition results and robust of system, this paper will focus on wearing glasses face recognition.Firstly,we proposed a Nonuniform Local Gabor Binary Pattern (NLGBP) method. Since NLGBP used Local Gabor Binary Pattern (LGBP) to represent face image, it could gain more feature information with multi-scale and multi-direction and it is robust to light changes. Meanwhile, however, during image division step of LGBP representation, LGBP approach adopted a simple uniform division strategy; which could not do well for wearing glasses problem. Therefore, this paper proposed a novel nonuiform division strategy which could treat different region with different influence role in the meantime gaining enough spatial information. So it is able to enhance the role of positive discriminable information meantime weaken the negative eyeglasses information for effectively improving the recognition results. NLGBP method although could deal with the wearing glasses problem in effect, it still try to construct a single optimal classifier to gain the best recognition rate.However, since the complicated of face recognition, the effect of single classifier is frequently inferior to mufti-classifiers,Based on NLGBP method, we proposed a Random Nonuniform Local Gabor Binary Pattern (RNLGBP) method. This method combined NLGBP method and random subspace method (RSM).It generated local histograms followed by random sampled from the local histograms to build multi-classifiers. Finally, it fused all of the multi-classifiers to gain a total classifier for the final recognition. RNLGBP method could take the advantages of RSM and NLGBP for improving the recognition results.Experiences shows that NLGBP and RNLGBP method both received a better recognition results compared to other method on dealing with eyeglasses problem.