Research On The Improvement Of Plate Image's Binarization

License Plate Recognition(LPR) system is a Computer Vision system focused especially on vehicle plates. It is one of the important subjects on Intelligent Traffic system based on Computer Vision and Patten Recognition. Among all the steps of a LPR system, the quality of plates binarization is very important because it influences directly the accuracy of characters' segmentation and recognition. In fact, a LPR system comes across many problems which will make the final recognition part fail.This paper improves the result of binarization in three aspects: Image preprocessing, Algorithm for binarization and Noise removal of binarized images.In the section of image preprocessing, this paper comes out a new algorithm named Smart Enhance which could differentiate the foreground from background effectively. Firstly, this algorithm maps the color space from RGB to HSV, then it extracts the plate image part through Sobel Edge Detection method and analysis on its vertical projection. Next, it classifies the V part of plate images into two parts with Ostu method. Finally, we scale the V value in order to increase the Contrast.In the section of binarization algorithm. This paper comes out a synthesized method from two algorithm: A local and dynamic method based on LOG arithmetic operator and a global method based on Ostu. The first method can enhance the edge data of characters while the second can run very fast and is effective under normal conditions.In the section of noise removal of binaried images, this paper analysizes and sums up the traits of the two kinds of noises in plates. It comes out alteratively two methods to remove the coresponding noises which are Detection of eight neighbour regions and Structual Analysis of the continuous regions.