Research On License Plate Character Recognition Based On Edge Template And Grid Density Feature

Along with our country market economy development and transformation, it has become inevitable to be into the automobile society, these changes bring the convenient and rapid society pace, at the same time it brings a series of urban road traffic management problems caused by the rush of motor vehicles. Traditional traffic management mode to shift under rapid development of information technology at present, vehicle license plate recognition system is the combination of computer vision, pattern recognition technology and digital image technology, which is one of the important research topics produced in the field of intelligent transportation applications. Vehicle license plate recognition system mainly includes five modules:the vehicle image acquisition, image preprocessing, license plate location, license plate character segmentation and recognition of single characters, this paper mainly studies the last module, that recognizes a single license plate character images segmented from the license plate image.As the license plate image is acquired from the actual environment, the spot and stain of license plate surface reflection can easily influence character recognition result, just a sort of method can’t satisfy the actual requirements. Because the letters and numbers accounted for the most proportion in the license plate characters, this paper focuses on recognition of the letters and numbers, at the same time in order to improve recognition efficiency, according to the license plate structural features and character features, this paper processes two-level classification recognition of license plate character. The first level classification based on the edge characteristic of the license plate character processes edge image K-nearest neighbors template matching, and determines whether they are similar character sets, if they are not in the sets, the recognition result will be output directly. If they are in the similar character sets then determined by the confidence coefficient, if the confidence coefficient is lesser than one regular value, to continue the second-level classification recognition, based on the differences of similar characters are mainly concentrated in the local, so this paper selects local feature area, combines with good invariance of grid density features weighted K-nearest neighbors matching algorithm to the deformation and rotation of the images to recognize. In order to reduce the amount of calculation, this paper adjusts the template library according to the order of the license plate character and chooses multiple difference character templates between the same kind of character.By studying the relevant theoretical knowledge and analyzing of the related algorithm, this paper tested license plate character images segmented from license plate images which were obtained under different circumstances, and used Visual C++ 6.0 as the software tool to achieve the algorithm programming, constantly studied license plate character recognition and tested experiment, analyzed of the various factors that may influence the recognition effect and reasons of false recognition, finally the improved algorithm was determined. According to the experimental results of study, which shows that based on the edge template and grid density features matching algorithm for license plate character can improve the recognition accuracy, and has done some try and effort from another Angle for the practical application of license plate recognition.