Research On New Type License Plate Recognition Based On Neural Network

License plate recognition technology involves the knowledge of computer vision technology, digital image processing, pattern recognition, artificial intelligence, and so on. Combined with these subjects'knowledge, a series of features information were obtained by analysis of the license plate. And on the basis of these features, the main steps of the license plate recognition were studied. Based on the previous researcher's work, we made an in-depth study on the recognition of "92" style and the new type license plate, and designed and implemented a license plate recognition software system.The license plate recognition system includs five parts, they are image acquisition, license plate location, characters segmentation, characters recognition and database management. In the part of image acquisition, there just made an introduction to the critical technologies. And following, some of the major image pre-processing techniques, such as image grayness, image binary, image gray enhancement and image edge detection, were introduced in detail. In the part of the license plate location, a license plate location algorithm based on means clustering and color segmentation was presented in HSL color space, and the experimental results proved that this method has a strong ability to anti-interference of the background, and has a high efficiency to the blue and yellow license plates'location. In the characters segmentation part, the license plate image was processed by slant correction and removed the frame firstly, and then it was divided accurately into single characters based on the plate priori knowledge of the fixed space among the characters and the vertical projection. In the paper, we described the character features in detail, and analysed some algorithms including coarse grid feature, histogram projection feature, feature-line pixel projection and Gabor wavelet feature, and also compared their advantages and disadvantages. Based on these features, an idea of fusion feature was inferred. In the stage of characters recognition, the characters on the license plate characters were divided into four categories, Chinese characters, English letters, numbers and symbols (including English letters and numbers), characters recognition algorithms (template matching and artificial neural networks) now commonly used were also analysed, based on these knowledge and combined with different features expression, the plate characters were recognized using four kinds of classifiers, and also there was an identification efficiency's analysis and comparison on different characteristics and classification methods. Finally, we designed the data management module of this system, it can provide users with data-enter, retrieval, query and other information functions. The license plate recognition system developed in the rechearch project has been finished, and the recoginition result is perfect. It can be used for automatic toll collection, traffic management, vehicle information query, etc. And it has a certain practical significance for the intelligent traffic control and management.