Research And Application Of Off-line Handwritten Digit Recognition

With the rapid development of information technology and multimedia technology today,the traditional paper text has been unable to meet the increasing needs of human beings, so weneed urgently to transform traditional paper text information into digital information. The printedpaper text is relatively mature, but the recognition of handwritten character due to its variousshapes is needed further research. The handwritten Arabia digit frequently used in postal code,test paper, bank notes, its special use determines the application requires high precision ofrecognition.This text researched and designed a complete SVM offline handwritten numeral recognitionalgorithm based on the accuracy rate of handwritten numeral recognition and speed ofrecognition. First of all, the handwritten digital image processing, according to the characteristicsof handwritten digits a normalization method is proposed，including location normalization andsize normalization. Secondly, in the extraction phase of handwritten numeral image features, putforward DTP characteristics based on the combination of CCH and DCCH. The preprocessedimage was divided into blocks, and then counted CCH and DTP statistical characteristics of eachsmall block, so got the image feature vector. Finally in the classification, this paper proposed amultistage classification method, used horizontal traversing times to divide the image samplesinto two, preliminary realize coarse classification; And then two multi-SVM classifiers wereconstructed for fine classification of image samples. The structure of SVM multi classifier used avoting strategy with a set threshold. If the votes exceed the threshold, the classification is correct;Or the sample will be put into another multi-SVM classifier.Finally, in the Matlab R2010software environment,60000training samples and10000testing samples of MNIST database were used in experiment. Experiments show that: the designof the handwritten digit recognition algorithm can obtain higher recognition rate and fasterrecognition speed, it has certain practical value.