Research Of The Technology Of License Plate Recognition

The intelligent transportation system(ITS) has become an important direction for the development of traffic management. The license plate recognition(LPR) system plays an important role as a part of the ITS, its wide application will help our traffic management the process of automation. In this dissertation, the three parts(license plate location, character-segmentation and character recognition) of the LPR system were studied. the main task includes the following:1. In view of the limitation of the template-matching algorithm, this dissertation finished the plate positioning combining the closing operation with the projection,what preprocessed the image first using closing operation and the background subtraction, and then the projection algorithm used to select candidate region. In accordance with the color characteristics of the license plate,the algorithm determined the correct region out the of the pseudo region. The positioning algorithm can make full use of the grayscale and color image characteristics, with good adaptability.2. This dissertation presented an binary method on the objective enhanced through the improvement of objective enhancement algorithm.Combining closing operation with the background subtraction,the algorithm can reduce noise such as license plate frame, so as to effectively improve the efficiency of the segmentation.3. This dissertation presented character-segmentation of license plate based on the enhancement of two-time binarization, in view of the shortcomings of only one-time binarization transformation in traditional character segmentation. The algorithm has twice binarization transformationon on the original license plate image.Combining with the feature of binary images to process the adhesion and fracture, this algorithm achieved good results.4. This dissertation recognizes Chinese characters in simple matching method,because of the dense strokes and the defect and the adhesion after the segmentation, and recognizes letters and numbers in feature-matching according to the simple strokes and the information can be most retained after refinement.