Research And Application Of Handwritten Digit Recognition

As an important branch of character recognition, handwritten digit recognitionhas a very wide application in the areas of postal communication, finance,transportation, Biomedical, etc.Relative to handwritten Chinese character recognition, the kind of handwrittendigit recognition is only ten, and its needing of precision of recognition is higher. Sothe studying on handwritten digit recognition is very important and meaningful.Handwritten digit recognition system involved the area of image processing andpattern recognition mainly. The main purpose of image processing is to preprocessand dismember the picture of handwritten digit, calculate the Features of the simplexcharacters, and then identify them with the method of pattern recognition. Traditionalmethods of pattern recognition such as Bayes decision of minimum error, templatematching and so on, are very suitable for the normal condition that the character shapeis regular, but will get the wrong results when it is deformation during the characterrecognition. Syntactic pattern recognition is capable of overcoming the situation ofthe character deformation perfectly, while its implementation is quite complex.Considering the two kinds of methods above, this paper propose a newcomprehensive recognition method based on the regular expression and templatematching. At first, by using regular expression the new method calculates a set ofcandidates made by all the possible results, and then find the precise result in the setof candidates with template matching. Regular expression settles the charactersdeformation and, template matching matches the similarity between the character tobe recognized and all the training data. A large number of experiments show that, thismethod is very easy to implement and has a good precision and practical value。According to the general process of handwritten digit recognition, this papercontains image preprocessing, character segmentation, feature extraction and selection,recognition methods including the experiment contrasts between Bayes decision ofminimum error, template matching, regular expression and the comprehensiverecognition method based on the regular expression and template matching, and thesound system of handwritten digit recognition at the end of paper.