Research On Character Recognition Based On Neural Network And Support Vector Machine

As the rapid development of economic, the character recognition technologyhas been paid more and more attention, especially the license plate recognition andbanknotes numbers recognition has become a research hotspot. The problem how torecognize the characters accurately and quickly, is a worthy of further research topic.Therefore, it has a certain theoretical significance and application value to study onthe character recognition technology. In this paper, it’s based on the results ofprevious studies and takes banknotes sample and license plate images as researchobject to realize the image preprocessing and character classification.Image preprocessing is the precondition of character recognition, this papermainly studied image acquisition, image filtering, image tilt correction, imagebinarization, etc. Using median filter technology to reduce the noise and otherexternal interference, and the Radon transform and bilinear interpolation algorithmare used in the image tilt correction. This pretreatment method not only highlightedthe characteristics of the characters, but also laid a foundation for the characterlocation and segmentation.Character segmentation is the foundation of character recognition. In this part,several common character segmentation methods were discussed in detail, and thenused the mathematical morphology method to realize character segmentation. Inorder to overcome the limitations of traditional image segmentation method, a newmethod based on HSI color space and improved K-means clustering algorithm isproposed. The provided method is used to extract the characters, the results showthat the method is effective and not influenced by noise and edge change, it providesgood foundation for the accurate identification of characters.Design an effective classifier is the key of character recognition. Artificialneural networks and support vector machine are the most widely used patternrecognition classification algorithms, this paper mainly studied the theoreticalknowledge of two kinds of recognition algorithm and its improved algorithm. Byrespectively using BP neural network and one-against-one support vector machinealgorithm to achieve the recognition of characters, which has achieved a goodrecognition rate and recognition speed. Comparing the simulation results, therecognition rate of two kinds of algorithms is roughly the same, while at the speed oftraining and recognition, support vector machine significantly faster than neuralnetwork method. The summary of the full text and the character recognition process,the furtherresearch and discussion of the character recognition technology have been shown inthe end of thesis.