Research And Achievement Of Tilt Correction And Character Recognition In License Plate Recognition System

As an important part of intelligent transportation, License Plate Recognition System (LPRS) iswidely used in our daily life, it plays an important role in modern traffic management, trafficdetection and vehicle inspection etc. The key technologies of LPRS are license plate location,license tilt correction, character segmentation and character recognition, the main contents of thispaper is tilted plates correction, character segmentation and character recognition.For the license plate tilt correction, this paper summarized the license plate tilt modes, puttedforward two different processes of correction and determined the best correction process by thesimulation test. Because the low quality images will bring interference to correction andsegmentation, so this paper studied the image preprocessing methods, it turned out that histogramequalization of the low quality of the image could be helpful to improve the success rate ofcorrection and segmentation. In order to improve the success rate of correction and reduce thecomputation time, firstly this paper studied Radon transform and then put forward fast Radontransform algorithm, the fast Radon transform can help to reduce the computation time. At last,combining the fast Radon transform with the rotating projection algorithm can improve the successrate of correction.In order to eliminate the non-plate area, this paper take use of the color characteristics of plate,proposed the HSV color space applied to detect the plate background and eliminate the non-platearea, this method can be applied to the color-undistorted plate. The projection detect transitionsmethod can be useful for the color-distorted plate. For the license plate separators and characteradhesions, this paper combined the priori knowledge with the vertical projection method forcharacter segmentation, the fusion algorithm can not only eliminate the plate delimiter symbol butalso can resolve the problem of character adhesions.For the classification of the character recognition, this paper studied the classificationalgorithms and chose the method of ANN to recognize the character. The Simulation results showthat BP neural network is better than template matching algorithm and the BP network is robust andit has practical value.