<title>The study of license plate location and character segmentation</title>

Abstract

ABSTRACT In order to improve the accuracy and precision of license plate location in license plate recognition system.The text introduces a license plate location algorithm with an adaptive threshold, which can achieve the exact location of the license plate based on an improved texture characteristic. The features of character segmentation are in the projection threshold which is able to change according to different images. Finally BP neural network is used to identify each character. Experimental results show that the method can achieve accurate license plate recognition and it has the following advantages such as accurate positioning, robustness and so on. Keywords: texture characteristics,license plate location,adaptive threshold,character segmentation,BP neural network. 1. INTRODUCTION License Plate Recognition (LPR) system[1,2,3] is an important component of the Intelligent Transportation System (ITS), which can automatically take photos of vehicles and identify the license plates. A complete license plate recognition system is composed of license plate location, character segmentation and character recognition , etc. License Plate Location[4,5] determines the location of license plates from the license plate images and separates license plates from the region segmentation. It is the prerequisite of character recognition, and the accuracy of location has a direct impact on character segmentation and character recognition accuracy. Most of the current license plate location algorithm is based on different features of license plates, including: (1) license plate character texture within the region [5,7] (2) geometric method[6] (3) color characteristics [3,7] (4) spectral analysis method [8] and so on. In practice, different positioning methods can be adopted according to the choice of different features. But, some of these methods are sensitive to the complex background and light, and others are slow in orientation .In order to overcome the shortcomings, this paper presents the license plate localization method based on improved texture features using adaptive threshold, to achieve the initial orientation of the license plate. In character segmentation, we use vertical projection. After and projection has been made, the threshold is improved to a certain degree of adaptive performance. Finally BP neural network is used to identify each character and to establish four networks, namely Chinese network, letters networks, digital networks and alphanumeric identification network. As our plates are unique, this classification is good for the rapid and accurate identification of license plates.