Automatic Vehicle License Plate Recognition System Based On Multiple Classifier Systems (MCS) Theory

Automatic Vehicle License Plate Recognition System(AVLPR-System) is one of the most important applications of Computer Vision and Pattern Recognition in Intelligent Transportant System(ITS). Based on the applications of Classifier Fusion Theory, improved algorithms are proposed in those three different modules of AVLPR-System.First, the error reduction for Simple Average (SA) and Weighted Average (WA) in the classifier outputs of the measurement level is analyzed.In the module of vehicle license plate detection, a novel classifier fusion-based algorithm is proposed. After locating candidate license plate regions, features of these regions are extracted for the optimal feature subset by exhaustive search strategy. Based on classifier fusion theory, Simple Average (SA) method and two Weighted Average (WA) methods are applied. Experimental results show that after reducing the dependency of features, SA method works better. The most three important features of the license plate regions are obtained in the experiment and our algorithm is robust in filtering out false plate regions.In the module of binarization of license plate, based on algorithm of CASDA (Cluster Algorithm Based on Spatial Distribution Analysis) an improved quick binarization algorithm I-CASDA are proposed to get rid of the influence of the illumination. Experimental results show that I-CASDA obtains better results than N.Otsu and CASDA, consuming time less than 0.01 second.In the module of character recognition, based on the step of feature extracting cascading classifier-based and SVM-based character recognition methods are proposed. Experimental results show that cascading classifier-based method spends 0.16 second, obtaining recognition rate of 98.17%; SVM-based method spends 0.37 second, obtaining recognition rate of 99.7%.