Study, Based On Bp Neural Network Character Recognition Technology

Nowadays, the pattern recognition technology widely applies in the artificial intelligence domain, like network search engine, document automatic sorting, biological features recognition, graph pattern recognition and so on. As a branch, graphic character's recognition has been aroused general interest, like the vehicles license plate, the zip code, the hand-written documents, printing documents, the webpage confirmation code recognition and so on exists to the pattern recognition demand. Because graphic character's recognition technology in many needs to carry on automatic diagnosis processing to the information the domain to have the enormous theory significance and the use value, we in view of the above demand, take the graphic character as the objects, based on the research of neural network character recognition and related imagery processing technology, summarizes a set comprehensively, the effective character recognition method. Meanwhile, seals the library which may supply to entrust with heavy responsibility to supply the application development use. This article will carry on the reorganization to above content's research process, mainly introduces next several aspects:1. Computer graph imagery processing method studyStudy Windows operating system's Graphics Device Interface plus (GDI+), analyze essential graphic object and image operating procedure which pattern recognition and the pretreatment needed.2. Research and realizing for the algorithm of target object localizationBecause the pattern recognition is carries on take the character as the unit, therefore need to locate the character object on the picture. We designed the marginal check method by to determine that character area coverage which presented on the picture, then extract the character object from the background images.3. Research and realizing for image optimization algorithmThe different origin picture recognition effect to be also infinitely varied, for promote recognition accuracy, needs to carry on the pretreatment to the goal character, including make gray, binarization, sharpen, de-noising, ascent adjustment operations and so on, this article has analyzed these operation committed step or provides idea for realizing.4. Research and realize the methods of neural network recognizionStudied the receptor state extraction process in pattern recognition as emphasis. Through the contrast discovery receptor state extraction process, combine with judgment by information entropy, had the very big enhancement in efficiency and compatibility compared to the traditional letter pixel characteristic extraction process.5. Research and realize the character recognition methodTake the BP neural network model as the foundation of character recognition method. Study the neural network part of open source project AForge Framework, uses it to realize the neural network foundation, realize the creation, training and character recognition by the framework.6. Carries on the seal to the core processing functionRealize the algorithm which before the code the stage research results, seals module that can be re-using. The module carries on the seal by the .NET assembly, may facilitate supplies C#, C++, VB.NET, etc.7. Design and realize the testingTo confirm the achievement of research, compiles the corresponding test order to carry on the confirmation, carries on the optimized improvement again according to the result to various stages algorithm.