Semi-supervised Classification Method Based On Manifold Learning And Its Application

With the rapid development of information technology and Internet, the past few decades have witnessed an exponential explosion in the availability of data from multiple sources and modalities. This has generated extraordinary advances on how to efficiently and effectively process massive amounts of complex high-dimensional data and extract valuable information, which are the common focus of several researchers from academic and applied math, pattern recognition and computer vision. A large amount of data is often showed obvious nonlinear characteristics, in order to solve this problem well, manifold learning algorithm are put forward. Manifold learning is an effective tool for data processing which can extract effectively streamlined information from the original high-dimensional data and found that low-dimensional essential attribute. But, most of the current manifold learning algorithms are unsupervised algorithms which algorithms do not use the prior information of the sample data. If they can get some samples of a priori information, this information can be used in the stage of training to improve the classification performance of the classifier, to promote ordinary learning algorithms get their semi-supervised algorithm.When dealing with high dimensional data samples, usually the first to reduce data dimension, the PCA algorithm is a commonly used high-dimensional data dimensionality reduction algorithms. Considering the PCA algorithm for failing to make full use of the prior information of sample, cause the dimension reduction efficiency is limited, and supervised learning and unsupervised learning many deficiencies in terms of the use of sample information, this paper proposes a discriminant analysis based on the manifold of a semi-supervised support vector machine algorithm. By defining based on manifold within the class of discrete degree and discrete degree between classes, give full play to the nature of the manifold discriminant analysis, so as to further improve a semi-supervised support vector machine, at the same time in the classification decision considering the boundary information, distribution characteristics of the sample data set and its local manifold structure, the method not only inherited the advantages of the traditional machine learning dimension reduction method, and makes the dimension reduction efficiency and classification accuracy of the algorithm has been greatly improved. Through the classic experiments on ORL facial database, verify the effectiveness of the proposed algorithm.