Research On Characters Segmentation And Recognition Based On Half-threshold

This paper makes an analysis and research of the application of OCR technology on mobile phone platform, and summarizes current OCR-related researches, including that character detection, segmentation and recognition. Our aim is to complete a mobile phone OCR system with high-performance character recognition in theory and application.Firstly, we analyze the limitations of mobile phone on image processing, and proposed an approach to guide image shot by character detection,which reduce the rejection rate of image,and improve efficiency of OCR system; Secondly, in the stage of character segmentation, we proposed a method to touch characters segmentation based on half-thresholding combined with character recognition; Thirdly, in the stage of character feature extraction, we proposed a method to complete feature extraction by classifying the character foreground pixels based on half-thresholding; Lastly, in the stage of recognition, we proposed a BP neural network which is based on genetic algorithm to recognize the characters.This paper asserts the validation of these method experiments on 50 english business cards, which proves 95.32% correct segmentation rate of touching characters, compared to projection method, ten percent of correctness higher than projection method; What's more, compared to the binarization of character foreground pixels method, our method shows some robustness on character feature. The BP neural network based on genetic algorithm obtains the ability of fast constringency and character identification. Finally, we implement recognition system for mobile phone Nokia N95 which is based on application and designed by using the Carbide C++, the image rejection rate is 6.25%, and the correct rate of character recognition is up to 97.35%.