The Research On Vehicle License Plate Location Algorithm

The thesis mainly carried on researching for vehicle license plate location algorithms in the process of the Vehicle License Plate Recognition, which is mostly applied in the scenes such as freeway charge gate and resident garage management.As a special computer vision system, the Vehicle License Plate Recognition System can capture a vehicle automatically and identify the numbers in the image. This system mainly composed of three key technologies, namely, vehicle license plate location, character segmentation and character recognition. Furthermore vehicle license plate location is more important key among the best, for location results will directly influence the correction ratio and recognition time eventual character recognition.However, there are so many difficulties in the vehicle license location, as new styles of cars, various figuration, complicated background, and so on. So the research for the vehicle license plate algorithm is still the hot project in the fields of academe these days.First, all kinds of location methods from other documents are systematically analyzed, and some key problems on vehicle license plate location are summed up.Second, this thesis put forward one algorithm which based on the structure feature license-plate location according to the complex background of vehicle images, the strong noise disturbance of vertical texture radiator conditions and so on. First, according to feature of license plate has high density, we obtained several candidate regions, and then we use four features to extract the license plate region from the candidate regions. And we give the threshold of license plate's structure features: the license plate region's ratio of width and height is range in [2,5], the license plate region's density is bigger than 0.25, the license plate region's gray-change time is range in [5,15].At last, this thesis put forward and proved number template matching license plate location method according to he strong noise disturbance of vertical texture radiator condition, because it's hard to decide the threshold, it cannot be applied.A great deal of experiments are done to contrast the methods presented in the thesis with former ones from documents, and the experimental results are objectively analyzed. The results show that these algorithms above mentioned give prominence to the correction ratio. According to 717 pictures, this method's correction ratio can reach 90.17%.