Research On Feature Extraction Method Based On Multi-manifold Learning In Face Recognition

Face recognition is one of the research hot spots inpattern recognition.The most critical task of face recognition is to extract the features of the image effectively.Therefore,For high-dimensional face image data,designing effective feature extraction method has proved to be the key to improve the performance of face recognition.However,the traditional methods of feature extraction are often affected by the noise of masks,illumination,expression,gesture and so on.for this reason,this paperutilize the idea of multi-manifold learning so as to devise and actualizeseveral feature extractionalgorithms for face recognitionwith strong discrimination.The main work of this paper is as follows:1?Robust Multi-manifold Discriminant Local Graph Embedding Algorithm Based on the Maximum Margin CriterionFor the existing multi-manifold face recognition algorithms,most of the original data with noise are directly used for processing,but the noisy data often have a negative impact on the accuracy of the algorithm.In order to solve the problem,a Robust Multi-manifold Discriminant Local Graph Embedding algorithm based on the Maximum Margin Criterion is proposed.Firstly,a denoising projection is introduced to process the original data iteratively,and the purer data is extracted.Secondly,the data imageisdivided into blocks and the multi-manifold model is established.Thirdly,combined with the idea of maximum margin criterion,the optimal projection matrix is sought to maximize the sample distances on different manifolds,while the sample distances on the same manifold are as small as possible.Finally,the distance from the testing manifolds to the training manifoldsis calculated for classification and identification.The outcome of experiment prove that,compared to the Multi-manifold Local Graph Embedding algorithm based on the Maximum Margin Criterion(MLGE/MMC),the classification recognition rate of the proposed algorithm is1.04%,1.28%and 2.13%higher respectivelyon ORL,Yale and FERET databases with noise and the classification effect has amarked improvement.2?Multi-manifold Discriminant Analysis Based on Kernel Sparse RepresentationAiming at the problem of non-linear separability in single sample face recognition,Amulti-manifold discriminant analysis based on kernel sparse representation algorithm is propsed.First,the data image is divided into blocks and incorporate the idea of multi-manifold.At the same time,the method of kernel sparse representation is used to depict the relationship between data points of manifoldsin the construction of the multi-manifold model so as to learn intra-manifold as well as inter-manifold graphs.Then,it find the best projections in each manifold space to maintain the characteristics of the intra-manifold graph while suppress the characteristics of the inter-manifold graph.Last,it calculate the distance from the test sample manifoldfor classification and identification.Experiments on Extended Yale B as well as CMUPIE datasets verify the proposed algorithm has better robust performancein illumination and occlusion changes than other algorithms.3?Sparse Constraint Non-negative Multi-manifold Graph Regular Incremental LearningIn order to solve the problem that non-negative matrix factorization lacks the ability modelling the data multi-manifold structure,a sparse constraint non-negative multi-manifold graph regular incremental learning algorithm is proposed.The algorithm introduces the multi-manifold learning and incremental learning into non-negative matrix factorization,and at the same time,sparse constraints are applied to the decomposed matrix,so that the sparsity of the decomposed data is improved,and the local representation ability of the image is stronger.in addition,the integration of incremental learning,after adding new training samples,avoids a large number of repeated operations,and improves the training efficiency of the algorithm.Experiments on public databasesverify the effectiveness of thismethodoverthe existing algorithms.4?Face Detection and Classification Recognition SystemBased on the visual studio platform,we design a face detection and classification recognition systemcombining with the proposed algorithms.Itsmajor functions include:(1)image import;(2)face detection;(3)face image processing(cutting,normalization,preservation);(4)feature extraction;(5)recognition.