Research On The Technology Of Face Recognition Based-on Locality Preserving Projection

Face recognition technology has been attached great importance to the researchers for its scientific significance and practical value in the past few years, and become the hotspot of current pattern recognition and artificial intelligence. It can be applied to security system, human ID management, teleconference, digital surveillance and so on. Face recognition normally be regarded as have three processes that are face detection, features extraction and pattern classification. In this paper, by systematically analyzing of relevant algorithms, researched on LPP (locality preserving projections) which is sort of manifold algorithm, center field LPP algorithm and constraints LPP is presented. We designed and implemented the prototype system of face recognition.The highlights and main contributions of the dissertation include:(1) Center field LPP algorithm is proposed. The LBP algorithm is simple and can be easily extended to extract the whole feature, we first obtain the global features using LBP, and then block non-uniformly the image, extracting the block which contains the eyes, mouth and other key organs of the center of the domain image, building nearest neighbor graph by the center field Euclidean distance between the image as a standard and finally get LPP algorithm for feature mapping feature space manifold. The algorithm can solve that the LPP algorithm can not be a good neighbor graph that represent the local manifold structure of space, and get a small number of feature dimension(2) Restraint LPP (constraint LPP) the algorithm is proposed. Consideration the LPP algorithm is unsupervised, with no use of information between the class samples, Unifiing supervised, semi-supervised and manifold learning algorithm, proposed constraint LPP algorithm. Firstly, create a marked and unmarked two neighbor graph, then according to different rules of assignment of weights mark neighbor graphs, put the marked weight as the unmarked weight constraint value, change the objective function, increased binding, and finally the feature map formation the new manifold feature space. Experimental results show that the algorithm can effectively use information between the class samples to improve the recognition rate.(3) Classification algorithm is proposed which called nearest neighbor distributed classifier. First we divide the training characteristic collection into many subsets, seek each training character subset average value and the variance, then judged in the subset whether closed with the test sample by the basis distributed thought. For the selected subset, use the nearest neighbor classifier to sort. This method will determine the distribution of ideas and thoughts of nearest neighbor can effectively reduce the calculation, and improve the recognition rate(4) A prototype system of face recognition with LPP algorithm is designed and implemented based on the idea of oriented object. We divide the system into four modules that is image preprocess, the LBP high-order statistical information, LPP feature space construction and face recognition.Accomplished constructing feature space by the different way,, get a different feature space. Through the face recognition experiments contrasts we proved the effectiveness of the above proposed algorithms.