Research On The Key Techniques Of License Plate Recognition

The vehicle license plate is ID card. With the rapid ascension of social economy and the rapid development of scientific technology, the number of vehicle, which is as people transport, is increasing yearly. And the trend of making false and illegally using others vehicle license plate is rising with the intelligent transportation technology development. Some people rack one's brains in scheming in making false license plate and illegally using others vehicle license plate in order to get selfish interests. These criminals spoiled our normally traffic order, and make some people suffer some unnecessary loss. It is very important practically for the research about the recognition of the vehicle license plate in order to severely punish the phenomenon of illegally using false and other's vehicle license, and driving illegal vehicles.Generally, the key techniques of license plate recognition is divided into three steps:license plate localization, character segmentation and character recognition. This article comprehensively apply Support Vector Machine(SVM), Fuzzy Neural Network, Rough Set, the Principal Component Analysis and so on, to study these three techniques,and meanwhile put forward some effective methods to improve them. Moreover it make use of the software of VC++6.0and Matlab to realize some key technology in the process of the license plate recognition. Specific content as follows:On the basis of comprehensively analyzing methods of the license plate location, this article raises the license plate localization method which is based on the fractal variance characteristics, combining with the mathematical morphology. Firstly, this method, which uses Ostu threshold method according to the image histogram, automatically select the threshold to realize the segmentation of plate area. Then, it erodes the image in the mathematical morphology erosion principle. Last, it realize the pseudo regional removed by extracting the connected region of the minimum variance of some part dimension, people can get the plate area which he interests. The above method can not only easily rule out the interference of large areas, but also play a good role to rule out small area.So it lays the foundation for the accurate division and recognition of the license plate character.This article adopt Multi-threshold fuzzy entropy segmentation algorithm in the character segmentation part. Firstly, this method combines points which are same in grayscale, and gets the average fuzzy entropy of the whole plate imagine. Then it selects the square, which the width equals to w=c-a+1. Constantly changing the r threshold, we can get many maximum threshold of the average fuzzy entropy, when the gray histogram appears multiple peaks. Finally we can use this threshold to get the segmentation of the license plate images which are located.The license plate recognition is the most important part in the whole link. The key of recognition is the choice of classifier and characters of the key feature extraction. This article creates a license plate character recognition optimization system for the sort of vehicle license plate's different characters.At first, it uses the improving of the BP neural network of license plate recognition for standard letter characters,and then it adopts the principal components analysis image reconstruction feature extraction and rough neural network combined, based on least square method support vector machine and p-harmonic model variation method of image restoration for fouling and blocking plate. Experiment results show that the method is effective, and solve the fouled character recognition in a way, it puts forward new solutions for false and illegally license plate. At the same time, this article discusses the similar character recognition.It improves the recognition precision and stability of the license plate recognition system with the above methods use.