Face Recogniton Based On Local Matches

Over the past few years, face recognition has been a hot research, as the demand in security and commercial applications growing rapidly. Many techniques of the facial features extraction have been proposed, and some of them have been succesfully put to use in life. LGBP(Local Gabor Binary pattern) is an excellent facial feature extration technique, which is qualified with excellent robustness to pose, illumination, expression, time variation and so on, but there still need to be further improved about illumination and time variation. The effection of time change is the same as the illumination change’s, the influence is caused by the albedo and shape of person’s face changed as the time. So in order to get a LGBP algorithm in all kinds of conditions of face changes are robust. The finally need to be sloved was the illumination change.This paper research on a better face recognition technique that robust to pose, illumination. expression, occlusion, time variation and the like as the goal, focusing on the LGBP of illumination changes carried out our research.First this paper in LGBP algorithm adopts an effective illumination preprocess method to weaken the illumination change to the influence of the features. Local normalization preprocess can effectively solve the illumination uneven influence by extracting the lighting irrelevant features fo the face image. This paper proposes an improved local normalization preprocess technology, we will be more effective in obtaining the attribute of light irrelevant and further improve LGBP the robustness to lighting change.Then put forward a kind of adaptive combined weighted method, this metod combines the entropy map weighted method that weights the testing image and adaptive weighted method that accord to the test images and contrast images automatically adjust the training weights, and makes the weights distribution tends to more rationalize, effectively improves the LGBP’s robustness under all kinds of changes.This paper proposed the improvement local normalization technique and adaptive combined weighted method, effectively improve the LGBP’s robustness to illumination change. For this paper gain a pose, expression, light, time change, robustness are good LGBP face recognition method offerd a great help.