Persian Vehicle License Plate Recognition Using Multiclass Adaboost

Abstract

Des pite the success of License Plate Recognition (LPR) methods in the past decades, this problem is quite a challenge due to the diversity of plate formats and multiform outdoor illumination conditions during image acquisition. This paper presents a real-time and robust method for Persian license plate location and recognition. The proposed method consists of four main steps namely (I) Plate localization (II) Normalization, (III) Character segmentation, and (IV) Optical character recognition. First, all plates of the grabbed image are located rapidly and accurately using morphological operation and AdaBoost. After that, plates are normalized, and if they are skewed, then they will be aligned. In next step, convolution of each plate and a predefined binary mask is calculated, and then characters are segmented based on the obtained information. Finally, SAMME is utilized to classify extracted Persian numbers, alphabets, and words. In comparison with other methods, this system achieves high accuracy in plate localization, segmentation, and recognition. The success rate of the proposed method is 96.93% for plate localization utilizing morphological operation and AdaBoost, 98.75% for character segmentation, and 94.5% for optical character recognition utilizing SAMME. Finally, the overall accuracy of the proposed method is examined 90.45%.