Research And Realization On The License Plate Character Recognition System Based On Neural Network

The license plate recognition system,developed in recent years,has an important application in the intelligent traffic surveillance and management.At the same time, based on the image and character recognition technology,the license plate character recognition system is a hot area of research in the application of pattern recognition at home and abroad.the license plate of a priori knowledge is rich,but in the context of complex,the license plate recognition is still more difficult.At present,most of the license plate recognition system is based on a simple scene and a single plate.the key technology of License plate character recognition system includes digital image processing,license plate location,license plate character segmentation and character recognition technology.In this paper,the system of license plate location and character segmentation,feature extraction,BP neural network classifier etc modules have had a more detailed research.Firstly,a collected color car image is changed to be a gray-scale image,and then local the position of the license plate based on Sobel edge detection and light _shade texture.Hough transform method that is rarely influenced by the noise and a continuous curve was Used to correct the tilted car plate which is made to be a tidy license plate.In addition,Otsu algorithm is used in vehicle license plate binarization which pre-process for segmentation of characters based on template matching algorithm.Secondly,put forward the feature extraction method of the gray-scale images of Chinese characters based on a pulse neural network PCNN,as well as feature extraction methods of numbers and letters based on th the characteristics of contour and 13 points grid.Moreover,discuss the e effects of a pulse neural network PCNN model parameters on the PCNN entropy Sequences of Chinese image,and ultimately determine the appropriate parameters of the PCNN for the system.Finally,designing two improved BP neural network classifiers to identify Chinese characters and other characters in license plate.in particular,the principle of BP neural network classifier,realization of BP algorithm and network structureare discussed.At thesame time,propose additional momentum factor and learning self-adjusting dynamic balance of the BP classifier method to improve the shortage of the standard BP neural network which is easy to fall into local minimum and has a slow convergence.In this paper,the output of BP network is expressed by the binary code.And the initial number of hidden layer node is determined by the empirical formula,which significantly reduces the number of iterations,reduces the size of network and increases the efficiency of the license plate character recognition Study shows that using pulse neural network PCNN to extract the feature of chinese character images has good robustness for translation,rotation,scale, distortion.In addition,BP neural network with good performance can meet real-time requirements of the license plate recognition in a complex environment,which has a certain theoretical and practical significance.