Research And Application Of License Plate Character Recognition In High Definition Images

With image processing and pattern recognition development, people pay more attention to thecharacter recognition. This thesis focuses on license plate character recognition in high definitionimages. After years of work by researchers, license plate recognition technology has already achieveda lot of breakthrough results. But the complicated traffic environment increases the difficulty oflicense plate character recognition and influences the accuracy of recognition. These environmentfactors include rapid illumination change, fast vehicle motion, the cleanliness of license plates, and soon. This thesis analyzes the existing license plate character recognition algorithms and difficulties ofsuch technology, and then implements a lot of new attempts. Main works of this thesis are as follows.Firstly, OCR technology and background of the license plate character recognition is studied. Aseries of pretreatment process are analyzed. Key research is license plate character coarse-to-finesegmentation algorithm based on vertical projection and priori knowledge. According to theapplication environment, this thesis compares and summarizes advantages and disadvantages of eachmethod.Secondly, the character feature extraction and selection method is studied and analyzed. Becauseof Chinese characters in license plate hard to recognize, this thesis proposes a novel feature extractionmethod based on orthogonal Gegenbauer moment. Experiments show that feature of Gegenbauermoment is better than that of Legendre moment.Thirdly, through discussing support vector machine method, we find the defect of such method,which emphasizes the largest classification interval, but not considers class divergence as small aspossible like Fisher discriminant analysis. This thesis adopts a new nonlinear classification algorithm-Fisher and comprehensive support vector classifier (FSVC), and then analyzes its multiclassclassification situation.Finally, the license plate character recognition algorithm has been applied to the automatic trafficaudit service system. Currently, traffic violation events are audited by manual work with high cost andlow efficiency. By applying this license plate character recognition algorithm in the system, the trafficaudit service system is faster and more automatic, which could help to reduce illegal behavior. Thisthesis also embeds the algorithm to a plate character recognition system on Android mobile platform.