Application Of Optimization Techniques To Face Recognition

With the rapid development of network digital and optimization technology, the knowledge of pattern recognition is used in increasing areas of practical problems. Pattern recognition problem is referred to discriminant classification problems of certain things or certain process labeled by training samples. Facial recognition technology, which is considered as the classic problems of the classification in pattern recognition, is mainly referred to train the data of training samples with category labels, and get a classification discriminant model to classify new unknown samples and then discriminate the affiliation of test samples. In the real world, the face image is easily affected by the changes of angle,expression, and posture. Matrix low-rank recovery decomposes the original matrix into a low rank matrix and a sparse matrix so that the separation of noise will be got. The main results of this dissertation are summarized as follow:In second chapter, motivated by nearest neighbor convex hull classifier, we put forward nearest neighbor convex hull classification algorithm based on the norm1 l of low-rank recovery. Firstly the algorithm aims at the separation of noise by decomposing the original training set into a low rank matrix and a sparse matrix and builds a new data dictionary by low-rank decomposition. Secondly, all kinds of points of the test samples and the new data dictionary convex hull model are set up respectively, and finally the classification is done by virtue of the distance of the unknown sample and the training set of convex hull model.On the basis of single specimen in the second chapter, the third chapter presents the norm1 l of low-rank recovery of multiple observation samples convex hull classification algorithm.First of all, this algorithm gains the plus form of the two matrix based on low-rank decomposition of the original training set samples and more observation of test set, and then sets up new data dictionary, and at last establishes convex hull of a new test set with more observation samples and all kinds of the convex hull of the model respectively, a new data dictionary by convex hull computation and more observation samples to all kinds of convex hull model of distance discriminant classification, makes the algorithm to the noise andsingular value has stronger robustness.In fourth chapter, the norm of low-rank recovery of nearest neighbor convex hull classifier proposed in the second chapter and the norm of low-rank recovery of multiple observation samples convex hull classification algorithm proposed in the third chapter are applied to face recognition problem. The data analysis vilifies the feasibility of two kinds of classifier in face recognition problem.