The Key Algorithms Research On License Plate Character Segmentation And Recognition

With the enhancement of people’s living standards, the popularity of the vehicle and presure oftraffic management are also increasing.In recent years, the study of car license plate recognition hasbecome a hot pot.This technology can be applied to traffic regulation, Overspeed detection, parkingfees management, car safety tracking and some other fields. Car plate recognition has four steps:plate location, tilt correction, character segmentation and character recognition.In car plate location, firstly this paper takes the color space conversion in order to exclude themajority of non-plate area, and then uses Prewitt to extract edge.Finally, detecting vertical andhorizontal edge, do a double-determination of the rectangle consisit of four edges. This methodimproves the accuracy in complex background. For license plate tilt correction,this paper analysisall the situation of tilt and conclude that vertical tilt correction has a greater impact on charactersegmentation,so takes the simple and fast algorithm Radon Convertion to correct horizontal tilt,anduses Rotating projection to correct vertical tilt.The hybrid method reduse the time and achieve agood pretreatment effect.In the third step of character segmentation,the paper firstly divides binaryimage of license palte into sevral rectangle area according to projection,and then selects thestandard characters.For suspicious rectangle aeras,uses connected domain and region growing to doa second cutting.For character recognition,this paper take the method of LS-SVM based on thewavelet kernel function,and then analyzes the feasibility of this kernal function,finally concludesthe advantage of wavelet kernel through the experimental data compared with thadition method andother kernal functiuons.The paper has analyzed and compared many classical algorithms in the four steps of car licenseplate recognition, takes high complexity algorithms in key points and simple and fast algorithms insome detail that do not affect the recognition rate.This paper reduses recognition time and alsoachieves with a high recognition rate compared with some tradition method.