Research On Single Sample Face Recogintion Algorithms Based On LBP And 2DPCA

Face recognition technology is based on human's facial features information, the process of face recognition technology is as follows. First, input facial image, to determine the existence of a human face, if exist, then output the following informations, all the faces' location and size and the relative position of facial organs. According to these information, extracted the facial features of each person and compared with the known face, so as to recognize the identity of each faces. However in many practical scenarios, face database can only collect one picture of face image as the training sample, but when in the case of a single training sample, many of the most classic face recognition algorithms' recognition rate is not high. Therefore, the research on the face recognition technology under the single sample has a broad market value. This paper studies of the single sample situation which may result the low face recognition rate and long recognition time.This paper mainly pursues research in three aspects:In view of the single sample face recognition problem, in order to fuse each local characteristics of facial image more effectively, proposed the improved adaptive weighted LBP face recognition methods. First, select the appropriate image division method, according to the different sub image classification performance, assigning different weights. Then according to the statistical knowledge put forward an idea of daptive weighted fusion and combining the LBP operator for face image detection and recognition.In view of low recognition rate of the adaptive weighted fusion LBP face recognition of single sample, proposed the single sample face recognition methods of fusing LGBP(Local Gabor Binary Pattern) and SIFT(Scale invariant Feature Transform) descriptor. SIFT descriptor has nothing to do with the location, size, and rotation of objects in the image change, and high stability in the micro perspective, noise and the scale. Gabor filter can obtain many important visual features. The method combined with both advantages and improve the problem of SIFT descriptor that has large amount of calculation, less extraction of practical ability to distinguish features. Then, to fuse extracttion features and choose the optimal feature vectors for face recognition.Base on the current popular two-dimensional principal component analysis and BP(Back Propagation) neural network algorithm and introduced more weight function, put forward single sample face recogination of twodirmensional Principal component analysis base on more weight function BP nueral network face recognition method of neural network. Compress the face image two times on the horizontal and vertical directions respectively. Then fuse weight distance in both directions respectively, and experiment in multiple classifier, compared and analyze with the results of the previous classical algorithms.To test samples respectively in ORL and Yale face database and prove the effectiveness of the proposed method and advantage.