Research Of The License Plate Location And Recognition Based On The Mathematical Morphology And Neural Network

License plate recognition (LPR) system is the core of the Intelligent Transportation Systems (ITS), which have been widely applied in various fields, like parking control, high toll station, electronic security and so on. In general, LPR system consists of the follow four parts:extraction of the plate region, tilt correction, character segmentation and character recognition. The paper contain plate license location, tilt correction and binarization, character segmentation and character recognition.As the plate location is concerned, the paper adopts the method based on the texture and shape feature of the plate. Using the principle that there is lots of edge in plate area, detect the edge by vertical edge operators. Then, sort these edge based on the method of by-point binary with the mean data. Finally, choose the plate area according to the aspect ratio of the plate and the approach of connected region search. The location rate could reach 90.6%.License images often have tilted after plate location, which would seriously affect the character segmentation accuracy. The paper proposes the method based on the Radon transform for plate tilt correction, which could obtain good correction results both for horizontal tilt and vertical shear tilt. At last, the method which combined with local threshold and global threshold is designed for license image binarization. Experimental results manifest that the method not only could effectively avoid the characters' strokes fracture, but also could fully reflect the characters'strokes structural information.The thesis proposes character segmentation method based on vertical projection and connected region correction; it is the step after the tilt correction. First, detect the border and connected region position after vertical projection of the license image. Then, correct the connected region position and extend character region combined with the prior knowledge until all the characters are segmented successfully. Experimental results show that the algorithm achieves good results for the license plate image which has been serious degradation like character adhesion.Character recognition is an important component of LPR system. This paper applies the clustering method based on wavelet transform coefficients for extraction the letter and digital feature and the block projection and clustering method based on wavelet packet transform coefficients for extraction the Chinese feature. In addition, a character recognition system using multiple networks based on generalized regression neural network (GRNN) is designed, which recognition rate could up to 100%,93.4% and 91.9% for Chinese character network, letter network and hybrid network.Testing the License plate from the real environment, the results show that the approach could be used to complex background of plate location, tilt correction, character segmentation and character recognition. Positioning method could not only accurately locate the license plate area, but also could eliminate interference factors in the area around the plate; tilt correction is the basis part for the character segmentation; the algorithm of segmentation has good segmentation results for plate images severely degraded; recognition method apply to not only alphanumeric, but also have higher recognition rate for Chinese character uniquely presented on the domestic license plate; and they are excellent in accuracy and robustness.