Face Recognition Method Research Based On2D Wavelet And Local Binary Patterns

Face recognition is widely research and application of a physiological feature recognition technology, after twenty years of development, have made progress in leaps and bounds, the fusion of modern pattern recognition, biometric technology, image engineering, computer intelligence, neural network and so on many disciplines of knowledge. The basic principle is to extract the training set image local feature information as the training pattern of sequence, and then extracted from the test set image local feature information as a test pattern sequence, through the classification to distinguish both information difference, finally gives the judgement result. So the local feature information extraction is the key of face recognition, pattern classification and recognition algorithm is good or bad will also affect the recognition efficiency.In the study of local two element model LBP (Loal Binary Pattern) basis, this paper proposes the use of two-dimensional wavelet analysis, can be the signal or image the time-frequency local feature analysis, using different filter, the difference compared to the local features of information, interested in finding the local information and features such as orientation, an input image from four different directions of decomposed into four sub image, various sub image to obtain the corresponding feature vector, and then select the appropriate sub-block size, on the block after images were local two element pattern encoding, because each sub image feature vector of high dimension features, using a linear subspace theory of feature vectors for dimensionality reduction transformation, better compression face image feature vector, faster, without reducing the recognition efficiency.In this paper, the method of face recognition, using the improved LBP operator, to obtain images of the more important local information, enhance and highlight the human face of the key points of interest, and the study of wavelet theory algorithm, proposes to the low frequency and high frequency are extracted feature vector, combined with the advantages of two algorithm theory.Based on Yale face database and the ORL face database. The experimental results indicate that, the proposed method compared with single method LBP recognition efficiency, such as in a variable light conditions and changes in the expression, but also can better identify, able to adapt to a certain intensity of illumination environment, is not sensitive to the change of expression.