The Implementation And Research Of License Plate Recognition System Based On SIFT

With people’s economic activities has becoming increasingly frequent, the demand for transportation has increased annually, which made vehicle violation, traffic accidents and other problems increased severely. At present, the emergence of vehicle violation detectors at the crossroads, speed detectors on the highway have effectively solved some of the traffic problems, but it is still in the experimental stage to give assistance to the police to deal with illegal parking at the roadside by handheld mobile client management system in real time and efficiently. Hence it is very significant to research the algorithm of license plate recognition according to the features of vehicle images taken by the handheld mobile device and develop preliminary simulation.Based on a simple and efficient algorithm steps, short operation time, and the ability to identify the inclined plate in a vertical view, this paper will realize algorithm research and the simulation system from following three aspects:the license plate location, character segmentation and character recognition.(1) Research on license plate location method. Because of the first character of our country’s license is Chinese character, whose feature of texture is rich, this thesis experiments with SIFT feature matching, SURF feature matching and the modified SIFT feature matching on three kinds of plates who are rotated ninety degrees, captured in night and blurry, and adopts the optimum one, the modified SIFT feature matching, to realize the Chinese characters matching. And then use RANSAC algorithm to eliminate the false matching; finally, calculate the affine transformation matrix to achieve license plate location.(2) The license plate character segmentation. Firstly, adopt the method of inverse transform to realize preliminary correction of the license plate. Then achieve re-correction by using Ostu binaryzation, Radon detection level tilt angle in turn. Use the connected domain labeling method to remove the plate frame, rivet and noise. At this time, the license plate is almost upright and rarely noise. Character segmentation is realized by using the vertical projection method simply and efficiently. Finally, the segmentation of characters are normalized.(3) Character recognition. The first Chinese character will be identified when the license plate be located. With the letters and numbers have little characteristics and some of them have the similar structure, this paper adopts matching method based on structure and the statistical characteristics to realize the recognition of numbers and letters. The second character of license plate is English letter, while the remaining is letter or figure. In order to improve the matching speed, the feature database of letters and the alphanumeric characters database are constructed. Matching and identifying the second character with the features of the 25 letters in the feature database of letters. And the next five characters are matched with the alphanumeric characters database.(4) Finally, the simulation system utilize the MATLAB to program and test it using 50 pictures of different motor car shotting by a mobile phone.