Research On Relationship Between Feature Extraction Of License Plate Numbers And Identification Of Incomplete Numbers

With the continuous progress of science and technology, the License Plate Recognition (LPR) system has also been developed rapidly. It has been widely applied to the system management of vehicle import registration in parking lot, Electronic Toll Collection and intelligent transportation illegal monitor, arid it is a very important part of the Intelligent Transportation System. The LPR is divided into4parts:the image preprocessing, the license plate location, the character segmentation and the character recognition. The main work and results of LPR are as follows:Firstly, the license plate image is preprocessed, including the gray processing, the image equalization and the binarization. Among them, the gray distribution of gray image is more uniform through the process of the image equalization, and it is good for the image threshold segmentation. Through the comparison of the two value image’s simulation result, it is proved that the license plate region is more obvious after image equalization, and it is also easier to locate the license plate.Secondly, the license plate location technique. The jump method is used to locate the horizon of license plate because the horizontal pixels of license plate region jump obviously, and the projection method is used to located the vertical position, what’s more the image is scanned from the1/3width to4/5width. The experiment result shows that it can not only improve the speed and the accuracy of location, but also eliminate the interference of the similar region.In the process of character segmentation, the conditional criterion is taken into account because of the situation of touching and broken characters. Then combined with the projection method and the maximum return type character segmentation algorithm, the characters are segmented, the process is that:firstly the image is vertical projected, each character width and the spaces between characters are recorded, and then the character is judged whether it is adhesion or fracture according to the proportion feature of character. Finally the accurate segmentation position is obtained by the maximum return type character segmentation algorithm. This method makes the character segmentation more accurate. Finally, during the character recognition stage, the characters are normalized at first, the character images are turned into same size of the template images, and then the grid features and stroke features are extracted, finally the characters are recognized by using the template matching method and feature matching method. The feature matching method is the method that it uses template matching method combined with the stroke features to recognize the characters. This method is more accurate, and when the characters are recognized, the area codes of characters should be compared at first, and through this method, some templates or some regions of a template can be exclude directly, unnecessary recognition comparisons can be reduced, and the speed of recognition is improved.