The Research Of License Plate Character Recognition Based On Neural Network Ensemble

Currently, the Intelligent Transport System (ITS) is a popular research subject in the field of world traffic. As an important constituent part of ITS, License Plate Recognition (LPR) plays a main role in the modern traffic control. In a manner of speaking that LPR is a very important and active research field in Pattern Recognition, no matter in the application background or the market demand, LPR both has broad space for development. With characteristics of great ability of self-recognizing and self-learning, fault tolerance and robustness etc., Neural Net Pattern Recognition is an important research area in the field of Pattern Recognition and widely applied into all kinds of areas.The starting point of this paper is to combine LPR technique and Neural Network organically and focuses on the research of LRP based on NNE with the use of integrated technology. The purpose is to solve the problem of license plate character recognition utilizing the idea of NNE, and then improve the whole recognition ability of the system to get higher recognition accuracy. Main works are listed as follows:1.Performing pre-operations on license plate images, such as Graying, smoothing and edge detection etc., and doing comparative analysis and research on the method of the license plate location and character segmentation; furthermore, extracting eigenvectors of organized Chinese characters, letters and figure characters with different feature-extraction methods so that providing a reliable basis for the stage of character recognition.2.In order to improve the learning ability of RBF NN, ameliorate the tradition RBF NN from two points. One is making Gaussian function and converse square root function to have linear combination to be one kind of new basis function, which makes the new basis function have stronger approach ability and generalization ability by adjusting coefficient dynamically; Another is improving the proposing the minus clustering algorithm of RBF NN. In order to make network have stronger learning ability, this paper gives a kind of algorithm named fast minus clustering algorithm, this algorithm can modify the cluster center based on the traditional RBF NN, which not only make the improved algorithm reduce complicated computation caused by updating density index but also amend the cluster center, thereby decreasing the bias of cluster center to cluster results; In another way, verifying feasibility and advantages of improved algorithm through approximating experiment to nonlinear functions and applying the improved RBF into license plate character recognition. In addition, through theory research and analysis on the emerging quantum neural network, the quantum neural network with multi-layer activation function is used into the license plate character recognition to obtain higher-accuracy recognition rate. Also, we can achieve similarly the license plate character recognition in this paper by using the BP network.3.Research and analysis the Neural Network Ensemble technology, in order to improve reliability of the whole system and correct recognition rate, this paper also gives a new kind of project about ensemble, which based on BP NN, RBF NN and Quantum NN. This project ameliorates traditional series-parallel connection ensemble method, innovate designs the NN Ensemble System which has two-stage parallel voting-mechanism. Do the Recognition Tests Experiment with plate images by Matlab7.0, and the result of experiment can prove that the efficiency and operability of the ensemble- mechanism which is given by this paper.