| Intelligent Transportation System, in which the License Plate Recognition is one ofthe key techniques, is the hot and difficult issue in the developtment of the moderntraffic management. At present, many License Plate Recognition systems have reachedthe pratical level, while correct recognition rate of some License Plate Recognitionsystem in an unideal circumstance needs to be improved. In this paper, some keymodules of the License Plate Recognition system are studied, including imagepreprocessing, license plate location, character segmentation and character recognitionand so on. The specific work is as follows:First of all, in allusion to the various types of noise caused by inside and outsideinterference when a license plate image is filmed in the open, this paper adoptes a newdenoising method, adaptive median filtering method. It also puts forward a new edgemethod (Log-Prewitt, Log-Sobel), which is used to sharpen image to make the imagemore smooth. According to the global threshold method, we select the appropriatethreshold, and then do binarization processing for gray image.Secondly, this paper combins with the mathematical morphology and colorcharacters to locat and segment a license plate, which can locate and segment thelicense plate region accurately.Thirdly, this paper adopts a special license plate character segmentation methodbased on the vertical projection and connected domain in mathematical morphology.The method can accurately segment each charater on the license plate region.Finally, this paper designes a neural network recognition of the License PlateRecognition, which aims at Chinese characters, letters, numbers and other characters. Itrecognizes license plate image by the standard BP network algorithm, comprehensivealgorithm which combined the additional momentum method with learning rate. Theresult of this expetiment shows that this method is better, and the correct recognitionrate can reach more than95%. |
PDF Full Text Request