The Research On Vehicle License Plate Recognition Technology Based On 1-SVM

As an important algorithm inspired by support vector machine (SVM), one-class SVM (1-SVM) is well applied to small sample clustering, nonlinear problem, novelty detection and so on. Machine Learning and Pattern Recognition in License Plate Automation Recognition System have wide application in Intelligent Transport, which receive extensive attention. Based on the research of 1 -SVM theory, License Plate Recognition (LPR) is of certain value and meaning.This research mainly consists of the following parts:(1) The Sequential Minimal Optimization (SMO) is proposed as the 1-SVM's training algorithm. To solve the problem of using one class classifier in the vehicle license plate locating, an improved clustering structure based on 1-SVM and hierarchical binary decision tree is proposed. By filtering out the non-target object step by step, the rate of classification is improved. In order to search optimum parameters, the Genetic Algorithm that has global optimal capability is used to adjust Kernel parameter in this paper.(2) The Hough method for slant correction and connected region method, projection, a priori knowledge for character segmentation are used in pretreatment and character segmentation of vehicle image. These methods overcome the defect that traditional method is affected by size of plate, solve problems of conglutination and declining characters, adopt to shorten the time of segmentation and advance the precision of classification.(3) The multi-sphere 1-SVM classifier (MSVDD) is proposed by expending the 1-SVM to multi-class model. When classifying, in order to ensure that boundaries are obtained using 1 -SVM tight enough for characterizing all classes in a given data set, introduce implicitly decomposition tactics, and propose MSVDD method in two different mathematical formulations in this paper. Use MSVDD in character recognition module of LPR, and gain the optimum scale of training data sets through samples gradually increasing experiments. The advantages and disadvantages of this method are analyzed to promote wider application.