License Plate Detection using Cascaded Classifier with Two-Phase Training and Testing

Abstract

License plate recognition (LPR) is a core infrastructure of most intelligent transport systems (ITS). Plate detection is the first and the most important step for license plate recognition. In this paper, several schemes are proposed based on cascade classifier to detect multi-plates in an image. Here, a two-phase testing approach is suggested to improve the efficiency of the system in tackling false alarm and improving the detection precision. Besides, a two-phase training with feedback is proposed by collecting negative data feedback to improve the training of the cascade classifier. Finally, a combined approach is also proposed by merging the two-phase testing and the two-phase training with feedback schemes. To analyze the efficiency of the proposed approaches, the experiments are conducted on the real-world images with the plates in different environments as well as the plates in two different languages with various size, complexity, illumination, multiple plates at front/rear of cars. The results show that the proposed schemes can improve the performance of the plate detection system.