License Plate Recognition Technology Research In Noise Environment

License plate recognition system is an important part of modern Intelligent Transportation System, which plays an important role in traffic monitoring, traffic management and many other fields. In recent years, LPR has become one of the important research subjects that computer vision and pattern recognition technology applied in intelligent transportation field. With the popularity of the car, the performance of vehicle plate recognition system is becoming increasingly important. LPR under ideal environment without noise can not meet the requirements of these days. Therefore, research on license plate recognition technology under noise environment is of great significant for improving recognition rate.This paper studies on license plate recognition closely around noisy environment. First of all, it needs to analyze license plate noise sources, image noise models and image noise denoising methods. Then it goes to the key technologies of license plate recognition which includes image preprocessing, license plate area segmentation, license plate character segment and recognition etc.This paper suggests a set of relatively integral vehicle plate recognition process and algorithms according to the study and research of traditional license plate recognition method. Firstly, the input vehicle plate image is preprocessed by image color space transformation, adaptive denoising, radiation correction procedures for avoiding the noise with a clear environment; then the image can be converted to a qualified format which requested by specified recognition demands; Secondly, the image of license plate is positioned base on edge extraction and morphological processing, and be segmented by color. For the edge extraction method, it has been chosen the sobel operator which not only insensitive to the noise but also extract the edge of license plate effectively by comparing noise sensitivity of different kinds of classic edge detection operator. And then characters are segmented, after image obtain tilt table correction and adaptive binary, the mobile window projection method is used for the characters segmentation, effectively reduce the noise influence. Finally, the processing of normalized and character stroke degree is used to character images, then a stronger robust correlation matching method is used for character recognition.The whole system is simulated by Matlab. The experimental results show that, under all kinds of strong single noise and strong mixture noise environment, the system of the design can achieve correct recognition rate above90%and effectively reduce strong noise interference, adapt to license plate recognition under noise environment, and has stronger robustness.