The Research On Character Recognition Of Vehicles' License Plates Based On Neural Net Integration System

Neural net pattern recognition is one of the important research directions in the field of pattern recognition lately. It has advantages of stronger ability of self-organiztion learning revising robust and non-linearity-handling,so neural net pattern recognition is widely used on all kinds of fields compared with traditional pattern recognition. And character recognition of vehicles' license plates is the important application of computer vision and pattern recognition in the intelligent transportation management and detection,the improved accuracy of vehicles' license plates character recognition system is required. Combining them is taken as the start of this article.In the article,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. I has finished followed work in article:1,Study the methods of image binarization,and present a kind of gray image binarization method and color image binarization based on SOFM.2, Choose the feature of rough grid and direction-line as the feature of character recognition,and improve the feature of rough grid . Optimize SOFM learning algorithm,then Study single neural net classifer based on Optimized SOFM learning algorithm which takes rough grid and direction-line as input .And use the single classifer on the figure letter and province recognition.3, Study the integration method of many inputs and many neural net .Then combine the Bp neural net classifer and SOFM neural net classifer to construct a integration neural net recognition system.4, Implement the correspondence arithmetic in Visual C++6.0 based on the theory research and construct a software of vehicles' license plates recognition system.The research in this article shows that using different feature for multi classifiers to recognize character can efficiently improve the robust and recognition rate of the recognition system compared with the single classifer. The method mentioned in the article has a certain reference value for the design of common the vehicles' license plates recognition system with random noise.