Research Of Car License Plate Recognition With Complicated Enviroment

With the rapid development of social economy,the number of cars on the road rises quickly,which causes many issues of the transportation management.An effective way to solve the problem of transportation management is adopting the Intelligent Transportation System(ITS)to reduce the cost of labor and improve the efficiency of the law enforcement.At present,the Intelligent Transportation System has been widely used in the bayonet of highway,parking lot and so on.The license plate recognition(LPR)system is an import ant part of the ITS.The license plate of the car is similar as the name of people,so it's very beneficial for us to recognize the license plate of the cars on the road.The LPR system is usually composed of license plate pre-processing,license plate detection,character segmentation and character recognition in sequence.The character segmentation is the middle step of the LPR system,which plays an important role in the system.In this thesis,we will first study the restoration of the blurred image,the grayscale algorithm and binarization algorithms for the image.Second we will study the character extraction algorithm of license plate,using the methods of regional growth and MSER respectively.Finally we will introduce the character segmentation algorithm based on the maximum inter class variance.In the section of preprocessing,at first,we introduce the method to restore the fuzzy image with the continuous blur kernel.The algorithm first models the fuzzy image and calculates the continuous blur kernel at each pixel of the image.The fuzzy image will be restored through the deconvolution of Richardson-Lucy at last.Second,we introduce the gray scale algorithm which adopts the weighted average method to highlight the green features of the image,and we also compare the results with different methods.Third,we introduce the hybrid binarization algorithm,which first calculates the global threshold of the image as the basis for reducing the noise of image,and then get the binary image with the local threshold method.In the section of character extraction,this thesis uses the methods of regional growth and the MSER algorithm to extract the characters on the license plate.The method of regional growth first selects one seed point,then clusters to obtain the block of the character.The method of MSER is an algorithm to extract the most stable region of the image.In this thesis,we first study the relationship among the gap of threshold,the maximum rate of change and the number of character blocks,so as to explore the optimal value of parameters.Next,it is also necessary to filter the characters extracted to obtain the appropriate one.Finally,this thesis compares and analyzes the results of the regional growth and MSER.In the section of character segmentation,we first extract one character of the license plate as the reference character,then eliminate the separation point of license plate to reduce the impact on subsequent steps.Then,we can get the ordinal number of the character by estimating the relationship of the reference character and these parathion point.Then we will establish the template of license plate.Finally,we will get the optimal combination of plate characters by the method of maximum inter class variance.This thesis focuses on the algorithm of deblurring,gray scale method,hybrid binarization algorithm,character extraction algorithm,and character segmentation algorithm.And the methods introduced in this thesis could be used widely.