Research On License Plate Recognition Based On Radial Basis Function Neural Network

License plate recognition which involves digital image processing, pattern recognition,artificial intelligence and computer vision techniques, is the key link to realize intelligenttraffic, and has important practical value in the field of intelligent transportation system.Based on the analysis of several kinds of commonly used license plate recognition method,this thesis combines with the research of the radial basis function neural network (RBFNN)and its algorithm, and designs a new license plate recognition system. The main constants ofthis thesis are arranged as follows:Firstly, the license plate location method and the color license plate locating method arestudied based on gray image and HIS space, respectively. Because these two methods processthe disadvantages such as long time consuming, easily affected by illumination, low locatingrate et al., we record the blue and white pixels of the plate by use of the scanning method andthe prior knowledge, and the license plate is reasonably located. The method improves thelocation speed and accuracy and is not easily affected by light.Secondly, the segmentation of license plate character is studied. The license plate region ispreprocessed with gray, and the edge detection is completed by the Canny operator. Then, theslant correction of license plate is carried out by the Radon transformation. Meanwhile, thebinarization is done by the threshold determined by the OTSU method. And then the plateframe and rivets are cleared away by mathematical morphology, and the license platecharacter is segmented and normalized by the projection method.Thirdly, the method of the license plate character recognition is studied. Contrastiveanalysis on license plate characters is taken by three methods, i.e. back propagation neuralnetwork (BP), radial basis function neural network (RBFNN) and generalized regressionneural network (GRNN)). In the same time, the selection of the RBFNN learning center andthe design of RBFNN and GRNN are investigated. The spread value of GRNN is optimizedby use of the Fruit Fly Optimization Algorithm (FOA). Finally, the graphical user interface (GUI) is developed, and the database of the licenseplate is set up. Therefore, we completed the design of the whole license plate recognitionsystem.