License Plate Recognition Technology

The vehicle license plate automatic recognition system is one of the important research subjects in the traffic department. Because of the variety of the types and conditions in different vehicle license plates, existing system can't solve all the problems. As the result, this subject is still intractable.The color images we acquired from web scanning or the CCD camera in highway station include vary time vary point of view and vary place. We take the following step after the locating the plate .Typical difficulties are the character blur, character destruction and character image stained because of the performance of the vidicon, the clean degree, illumination conditions and the movement of the vehicle and so on..The whole system includes preprocessing segmentation and characters recognition. In the vehicle images preprocession, the paper use a binarization method base on spatial distribution, and realize the unifying of background black color. Then the paper presents a method used Radon transform to pick-up oblique rate and truly remedy the vehicle license plate image. Finally the paper bring a bilevel image horizon projection method to cut characters' thereabout disturb. In the segmentation the paper makes the best of the license plate normal apriority knowledge, depending on vertical projection to segment single-character. Finally we use BP neural network to recognize the characters.Testing more than 200 representative vehicle license plate images, we can achieve the accuracy of 98% in correcting incline rate, 95% in segmentation and 90% in recognition. The used technique is proved successful and feasible.