Research On Segmentation Algorithm Based On Statistical Classification For Mixed Characters

With the popularity of the information and computer technology, it is anurgent problem to transform the original information from variety media intobinary form that can be processed and transported by computer.OCR system is an important technology aiming at transforming themanuscript characters into computer ones automatically and is used widely invariety fields. It has been produced large benefit to society and economy.At present, most OCR systems are based on single character recognition.Character segmentation is one of the key factors that directly affect the rightrecognition rate of OCR systems.In this paper, a method for mixed characters segmentation based onstatistical classification is presented for the Chinese and English Charactersmixed image. Firstly, character is segmented with projection method and then isclassified with Bayes classifier. Secondly, with the help of recognitiontechnology, character is segmented precisely, and is judged whether it ispunctuation or part of a character;lastly the parts of character are merged andthe result is confirmed after the recognition. To judge and then extract theconglutinate characters, a heuristic outline tracking method is adopted in thispaper. For the character with underline, the underline is extracted and wiped off,and the recognition module is executed to segment this kind of special character.Experiments show the segmentation effect is satisfying for mixed arranging ofChinese and English characters in this method. The right recognition rate andsegmentation rate can reach 99%, which has been tested in the experiments.