Research and Realization of License Plate Character Recognition Algorithm Based on SVM
|School||Nanjing University of Posts and Telecommunications|
|Course||Signal and Information Processing|
|Keywords||License Plate Location Image preprocessing Character Segmentation Character Recognition Support Vector Machine|
The support vector machine (SVM) as an important theory of machine learning and pattern recognition, and has been widely applied in the field of learning to solve the small sample clustering, nonlinear problem, outlier detection. Automatic license plate recognition system is an important application of machine learning and pattern recognition technology in the field of intelligent transportation by widespread attention. Therefore, based on the license plate character recognition support vector machine technology, has a certain theoretical significance and practical value. License plate recognition system, mainly by the license plate pretreatment with positioning plate character segmentation, license plate character recognition of three parts. In this paper, the acquisition of the license plate of the actual scene images, research and design a license plate recognition software experimental system. The specific study contains the following: license plate pretreatment module using gray-scale, contrast enhancement, median filter, Canny edge detection, binary and other methods; positioning module, using the line scan mode and vertical projection and other means, to determine the upper and lower bounds of the license plate and the left and right borders, and laid the foundation for the subsequent character segmentation. 2 in plate character segmentation module, used mainly to Hough transform based license plate tilt correction, can effectively detect the tilt angle of the license plate, and the correction processing is performed in real time; for the correction of the plate after the division processing, combining the character inherent features and geometry, the use of vertical projection, the threshold delimitation method to determine the character of the division boundary, to overcome the adhesion of characters, character segmentation errors shortcomings fracture case, effectively improving the accuracy of the character segmentation. Character recognition by the divided character normalization process, to extract the license plate characters, using a combination of the sequential minimal optimization algorithm SVM classification prediction and cross-validation small sample optimization parameters; article also done sample increments experiments, the experimental in the the limited samples under the conditions of the optimal sample size; license plate character recognition errors, supervised learning, re-learning training and modeling and then adding the sample library, can improve system recognition rate. Finally, combined with the laboratory's existing resources and experience of their predecessors, designed an experimental system of license plate recognition software, and interface functions and database design process.