Imaproved Optical Character Recognition based Automatic Number Plate Recognition

Abstract

In recent years, there is significant development in intelligent transportation which grabbed more attention. An automated, fast, accurate and robust vehicle license plate recognition system has become necessary for traffic control and law enforcement of traffic regulation such as speeding, red light infringements, fatigue offenses by commercial drivers and other illegal behaviors. The Automatic Number Plate Recognition system (ANPR) is an Artificial Intelligence system which can extract, identify and recognize the vehicle registration plate number from an input image using Optical Character Recognition (OCR) technique. The license plate recognition system can be implemented over a wide platform like public parking, road crossings, highway tax collection and surveillance, district vehicle management system, and so on. Although currently there are many existing Automatic License Plate recognition systems, the research and development of algorithms have never stopped. The thesis is dedicated on the technique of Automatic License Plate Recognition System. A blended algorithm for recognition of license plate is proposed and is compared with existing methods. The whole system can be categorized under three modules, namely License Plate Localization, Plate Character Segmentation, and Plate Character Recognition. Detection of foreground object is the first relevant step of the system, and background subtraction techniques assist in fulfilling this step. The paper puts forward an novel algorithm for License Plate Localization including preprocessing techniques such as noise removal, image smoothening, color space conversion and plate edge detection. The algorithm is optimized to overcome illumination variations, shadows, background clutter and camouflage. The recognition scheme combines adaptive and iterative thresholding with a template matching algorithm and intelligent training of features extracted from characters using Artificial Neural Network. The main advantage of this system is increased recognition rate, robustness in recognition, computational efficiency and real time capability. The system is simulated under MatLab software platform. Simulation is carried out on 300 national and international monitoring vehicle images including cars, motorcycles and other vehicles containing number plates and the results obtained justifies the main requirement and purpose of detection and recognition of vehicle license plates.