Fingerprint Classification Based On Svm Research

Fingerprint recognition technology is one of the most significant and reliable identification methods as it is unique and invariable. With the development of computer science, fingerprint recognition has extensive outlook servicing people. It is widely used in the field of international trade, criminal act and a course of justice. So the research of an automatic fingerprint identification system (AFIS) has important technical value and practical value.There are many algorithms in field of fingerprint verification, but they still have some shortages to be researched. This paper studies the building of the fingerprint image database, fingerprint preprocessing, feature abstraction, and fingerprint classification. From all of this, fingerprint classification is the keystone of this paper. There are two algorithms proposed. The main work of this dissertation is as follows:1. In the fingerprint acquisition process, this paper built fingerprint image template database with 4000 images in it. Based on the interface function, this database saves the path of images and features as its data member. A novel multi-stage index method was proposed combining with the algorithm of classification and matching.2. After comparing the algorithm of others, this paper chose a series of reliable preprocessing and feature abstraction methods for its fingerprint classification.3. Using Support Vector Machine theory, the main research work of this paper is fingerprint classification. In the beginning, based on the rule of 1-against-rest this paper designed a two-stage classification system using five SVM classifiers. Program realized it. The result of the experiment showed the system has perfect generalization ability and was feasible in practice. Though the original work can greatly improved the accuracy rate, because of the number of SVM classifier is too large, which decrease the rate of operation. In the following work, this paper used binary tree theory to make a decision and only used SVM classifiers in the second final node. So the number of SVM has reduced to three and the rate of operation has improved greatly.