Recognition Of Merged Text-based CAPTCHA

The 21^st century is the era of information explosion,Internet technology develops rapidly and brings great convenience to people's life,at the same time all kinds of network resources abuse problem has caused the attention of researchers.CAPTCHA arises at the historic moment,has been applied to each big website,in order to prevent resource being occupied by a malicious computer program and protect the privacy of the information.For the recognition research of CAPTCHA,not only can point out the holes of CAPTCHA to prevent automated attacks,virtually also promoted the development of digital image processing,pattern recognition and machine learning,etc.This paper mainly research the recognition of merged character CAPTCHA,the characters are merged in this kind of CAPTCHA,some are merged simply,some are complicated.Six websites' CAPTCHAs that are used for login or registration are picked out for research,including bank of communications,CSDN,Zhihu,Sina email,Sina WeiBo and NetEase,the paper puts forward the recognition algorithms according to their respective characteristics.The main research work and results in the paper are as follows:1?According to the problem of the merged and slant characters in CAPTCHA of bank of communication,this paper proposes a improved projection segmentation algorithm,which join the dynamic rotation in the projection segmentation algorithm and improve the accuracy of segmentation point.According to the problem of segmentation of the tight merge in CSDN,this paper combine the projection segmentation algorithm and widths and pixels.The segmentation rate can reach nearly 100%.2?For the fragmentary contour of Zhihu CAPTCHA after binarization,this paper proposes a new algorithm for repairing contour lines,which searches the matching points depending the region and scale.In addition,connected area segmentation is improved when combining fragment of characters.Aiming at cracking the register CAPTCHA of Sina emails,this paper mainly depends on the judgment on convexity and concave of noise and characters,in order to denoising after extracting blank parts.3?According to the problem of the distorted,slant and merged CAPTCHA of Sina WeiBo,this paper put forward a improved drop fall algorithm,which combine the drop fall algorithm,rotating jam algorithm and vertical projection algorithm and solve the problem of slant and successfully divide the merged characters.Combine the CFS,vertical projection and drop fall to recognize the NetEase CAPTCHA,mainly according to the respective division characteristics,gradually and successfully separate the each character from the image.