Vehicle Plate Automatic System Research Based On Neural Network

The Vehicle License Plate automatic Recognition System (VLPR) is one of the important research subjects in the traffic department. It has expansive apply forground in electron policeman, freeway automatism charge and intellective park management. The images come from natural scene exist many stochastic interference of external environment. So I must try to reduce external influence. The difficulty in the License Plate Recognition is skew emendation, and blur recognition. The paper presents some means for these problem, so that the system can efficiently accomplish character recognition.The primary direction of my research is the study of vehicle license plate recognition system, including the pretreatment of license plate image, license plate character segment and character recognition.of license plate image is composed of reduce influence, image buildup, image binarization, background hue unity, skew emendation and so on. The difficulty in the pretreatment is skew emendation. Meantime, I present to use improved hough algorithm so that rectify a skew detection of license plate image. The paper presents a pretreatment method able to solve the problem that the image contrast degree is too lower to our eyes because of unconstrained illumination conditions.The difficulties in license plate image segment is the effects of any disturbance of yawp, character blur, character conglutination and character rupture etc The paper presents some methods that adopt prior knowledge of the license plate, perpendicular projection and so on.. It makes full use of the feature of the image and back scan. The priori information of license plate does well with the character conglutination and character rupture and presents a nice pre-segment.Typical difficulties in in license plate image character recognition is the character blur, character destruction .character image stained because the effects of vidicon's performance, license plate's neatness, illumination etc. So arithmetic must have good adaptability. The system use BP neural pattern recognition. The BP neural pattern recognition is considered as the recognition method and vehicles' license plates is taken as recognition object, which is to improve the recognition rate of the character recognition of vehicles' license plates under noising circumstance. In the end I get a good license plate recognition result.Testing 100 representative vehicle license plates, I can achieve the accuracy of98% in segmentation and 95% in recognition.