Research Of License Plate Recognition Arithmetic Based On Wavelet Transformation And Neural Network

License plate Recognition integrates computer vision technique and pattern recognition technique. It plays a key role in charging automatically on bridge and at the crossing, realizing automatic management at parking lots, registering the actual time and calculating the flow of vehicle coming and going at transportation crossing, identifying and holding up robbed vehicles and those involved in troubles and criminals.The thesis mainly covers such points as follows:Image preprocessing: Traditional methods of image preprocessing like histogram diagram balanced and highpass filtering have such limitations as enlargement of undesired parts and introduction of new undesired parts. Considering these shortcomings, the thesis brings about a method of enhancing digital image based on wavelet transformation, which attains image multiresolution grade information via wavelet transformation, improves the image contrast thought modifying image grade information coefficient under different resolution ratio scales.License plate location: Considering the vehicle images under different backgrounds and illumination conditions, this thesis brings forward the calculation method of locating license plate with improved Sobel arithmetic operators. It involves initial processing of pre-treated license plate images by applying improved Sobel, specifying the foreground with threshold, using the horizontal gradient algorithm to get the horizontal projection image according as the characteristic of rich license plate regional veins, then using the opening and closing operations of the mathematical morphology to do vertical location of the license plate and locating the license plate finally.Character segmentation: Considering the influence of noise in the area of vehicle plates after location and unavoidable location errors, the thesis brings about the character segment arithmetic based on the integration of wavelet's removing noise and vertical projection. The method involves eliminating the influence of noise via wavelet, character segment processing images with noise removed by vertical projection technology, getting single numbers, letters and character and unitizing the license plate.Character Identification: With license plate character as the identification object, BPNN model identification method researches on the license plate identification problem in the situation of interference. The method has turned out to be effective.