Research On The Key Techniques Of License Plate Recognition

The research and development of car License Plate Recognition is the linchpin of the Intelligent Transport System, and it plays an important role in the ITS, such as information collection, car management, automatic toll collection and so on. This paper mainly discusses several portion of the LPR system.In the aspect of plate location, an approach based AdaBoost and Harr-Like feature is discussed . First the Harr-Like feature of both true and false plate images and their edge detecting results is computed to train a large number of weak classifiers. Then some appropriate weak classifiers are boosted to be a strong classifier through AdaBoost algorithm. The strong classifier is used to detect the plate area of the image. Finally a simple cascade network based AdaBoost is constructed for attempt. The experiment proves that this approach can locate the plate rapidly and efficiently.To fix the geometric distortion of the located plate, the Hough Transform is applied to get the tilt angle of plate, then geometric transform such as image rotation, and bilinear interpolation are used to reduce the distortion.The single character is picked up from the plate according the transcendent knowledge of structure and projection feature of car plate. To repair the possible broken strokes, a measure that is based upon either run-length direction or skeleton extraction is adopted and then the broken stokes are connected.In conclusion, this article mainly discusses four portion of LPR system, such as car plate location, geometric distortion fix, characters pick-up and strokes repair of single character. Each of aforesaid is implemented with computer program, and the result is acceptable.