Research On License Plate Recognition Based On Image Processing

License plate automatic recognition technology is an important component of modern intelligent transportation, as the city's rapid development and the growing license of motor vehicles, there is a higher attention on the vehicle license plate recognition technology (hereinafter referred to as license plate recognition).License Plate Recognition has a wealth of prior knowledge and become an important foundation for the field of pattern recognition, but in the complex background and different weather and illumination conditions, it is still a hot issue that recognize the characters of license plate accurately. The technology in papers including four aspects which were the license plate image preprocessing, license plate location, segmentation and recognition. According to the advantages and disadvantages of traditional algorithms, obtained by analyzing the situation, the algorithm is improved, and the results about the new algorithm are gived.First, do some preprocessor which include image enhancement, binarization and slant correction. Bring forward a histogram equalization algorithm which bases on the local twin peaks adaptive selection to enhance the contrast of image, the mentality of this method is: select Threshold to cut up background target image, then weighted stack them to get the enhanced image. In view of the major noise in License plate images is impulse noise, bring forward a adapting median filtering algorithm to denoising. According to the histogram of enhanced images, work out the threshold for binarization. Toward oblique images, calculated oblique angle using Hough, and then rotate the License plate images correction which are in different tilt type.Secondly, in theory, classify and sort the license plates for our country, bring forward a segmentation algorithm which bases on the license plate character edge detection and its color features, use Prewitt detect vertical edge of image, scan and statistics edge point to locate the candidate areas of License plates, use FCM method confirm the true area of License plate. According to the areas of the License plates, complete a single character segmentation using the vertical projection that combined with prior knowledge of the license plate character segmentation. Finally, design the neural network classifiers in order to train and recognize the character, letters and licenses. Through the analysis of character features, confirm the input layer, hidden layer and the License of output layer units. The results show that the convergence rate of the net work is faster, while not easy to fall into local minimum, reducing the License of iterations and the network size and increasing the efficiency of License character recognition at the same time.The results show that:in this paper, recognize the actual filming of the License plate image with the recognition rate more than 95%,using the method which is image processing techniques combining neural networks.