Research On License Plate Recognition Technology Based On Image Processing And Support Vector Machine

With the expansion of urbanization in our country, the number of urban population andvehicle ownership in major large cities is increasing, which brings urban traiffc congestion.The rise of intelligent transport system (ITS) alleviates the increasingly serious traiffcpressure to some extent, and promotes the development of transportation science information,intelligent and socialization. Automatic license plate recognition system as a core of ITS，which has been widely used in many ifelds. It is a typical license plate recognition system thatincludes image preprocessing, license plate location, character segmentation and characterrecognition. Algorithms of the main part of the system, including image preprocessing,license plate location and character recognition, is respectively provided and researched inthis thesis.(1)License plate image preprocessing. Image preprocessing mainly includes red greenblue (RGB) image grayscale transformation, nonlinear gray stretch, histogram normalizationand binarization. After the image preprocessing that reduces the complexity background ofthe license plate image, improves the accuracy of license plate recognition and saves thecomputation time and storage space.(2)License plate location. It adopts a method of license plate location based on thetop-hat transformation and mathematical morphology connected posture analysis in this thesis.After the top-hat transformation, which gets rid of the effect of the background. Thenapplying the improved Sobel operator detects license plate edge to get license plate candidateareas. According to the size of the plate itself characteristic, using mathematical morphologyconnected posture analysis to iflter the noisy for candidate license plate areas, and using4-neighborhood with color symbology to sign. At the same time, the license plate is locatedby using horizontal and vertical scans. The simulation results show that localization method isnot only feasible and effective, but good performance has been achieved.(3)License plate character recognition. Through the collection of samples of licenseplate characters are normalized to use wide-gridding character, and use genetic algorithm (GA)to get the optimized parameters of radial basis function (RBF) kernel function under thecondition of small sample, then use the optimal parameters of RBF kernel function of supportvector machine (SVM) classiifcation prediction to improve the recognition rate of thecharacters effectively. Finally, using MATLAB and Lib SVM toolkit to test and veriyf themethod given by this thesis where GA is adopted to optimize the recognition of the license plate character of SVM，and compare the character recognition effect of SVM and BP neuralnetwork.