The Study Of Support Vector Machine And Fingerprint Classification Algorithm

Support Vector Machine (SVM) is a machine learning method based on the Structural Risk Minimization principle and VC dimension of statistical learning theory. By using kernel functions, SVM eliminates the problem of "curse of dimensionality" in high dimensional space and has better generalization performance than traditional statistical based learning methods. As to its application to multiclass classification problems, most of the methods currently used are based on combining many binary SVM classifiers to build a multicalss classifier. In this paper, a new method is proposed. Unlike the method mentioned above, unsupervised single-class SVM classifiers instead of supervised binary SVM classifiers are used as the basis for multiclass classifier construction and the single class classifiers are combined in three ways to provide a multiclass classification result. The total number of subclassifiers in our new method linearly scales with the number of classes and the size of each subclassifier is smaller than that of the correspondent binary SVM subclassifier used in the original method. This new method is used to solve the five-class fingerprint classification problem in which fingerprints are classified into whorl, left loop, right loop, arch and tented arch and satisfactory results are got hi the experiments. At the same time, Gabor filter is used to extract the global feature of fingerprints from four directions and a graphic user interface is designed for this application.