Research On Algorithm Of License Plate Recognation

With the increment of the amount of vehicles, the traffic situation in urban area is exacerbating. More attention is paid on how to manage the traffic effectively. The vehicle license plate recognition system is a newly developed intelligent transportation management system based on the technology of digital image processing and pattern recognition. It has been widely applied in areas such as toll payment and traffic management; it can help improve efficiency and reduce manpower, which makes it a both economical and practical system. Therefore, the vehicle license plate recognition system has great importance both economically and practically. The vehicle license plate recognition system mainly consists of three parts: license plate location, character segmentation and character recognition. The algorithm of each part is studied in this dissertation:(1) For the license plate location problem, a license plate location algorithm of color plate based on Harris corner detection is proposed, considering the characters of license plate have enough corners with the advantage of HSI color space. The image is first converted from RGB color space to HSI color space and filtered; Detecting corners of saturation component is then carried out according to hue component; Finally, the area of license plate is searched by rough and accurate location based on the characteristic of license plate,. This method can locate license plate correctly in various conditions and the cost of system is lower than traditional methods.(2) For the character segmentation, firstly, the angle of oblique plate is calculated based on the information of corners and the plate is then corrected; Secondly, the plate information is converted to binary using a fixed threshold, hue and saturation components are also taken into consideration; Moreover, the effect of impurity is weakened by using horizontal and vertical integral projection and the characters are segmented; Finally, characters are smoothed by using mathematical morphology. Experimental results show that this method can correct license plate successfully in any angle with low system cost, segment characters exactly and renovate it effectively.(3) In the section of character recognition, the contour feature of the segmented characters is extracted. Four classifiers are constructed based on the characteristic of plate. Then single character is trained and predicted by SVM with RBF kernel function. One-Against-One classification method is adopted in this step. The selection of RBF kernel parameters and penalty factor C is described. The experimental results show that this method has excellent recognition ability even in the small-sample case.