Recognition Technology Research Of Distorted And Merged Text-based CAPTCHA

There have many network problems,such as batch register emails by network robot,forums irrigation,malicious brush votes,brute force passwords etc.It not only causes a waste of resources,affects users normally use but also takes security issues in user accounts.Twisted adhesion CAPTCHA can increase the difficulty of machine recognition.The recognition research of this CAPTCHA class not only be designed a more security and facility CAPTCHA from the perspective of anti-identification,but also can discover the security vulnerabilities of the CAPTCHA,so as to improve the security of the internet.In this paper,the CAPTCHA image pre-processing technology,and the CAPTCHA identification theory in-depth study.Because twisted adhesion CAPTCHA outliers are more and more serious noise,based on the traditional identification methods,etc.shape the context for this type of codes deformation invariance problem processing ability is poor,to solve this problem,this paper based on relative shape context and point pattern recognition methods,and Net Ease mailbox and CSDN existing verification code identification,effectively solve the Net Ease mailbox and CSDN character codes adhesion problems.In addition,the traditional shape context algorithm there are character codes,such as the shape and contour noise sensitive,accurate and describe the shape features are not calculated cancroids of the complex time-consuming and other defects.In this regard,we propose the SIFT(Scale-invariant feature transform)feature and an improved combination of shape context algorithm using SIFT descriptors acquisition codes shape structure information,using some of the least squares algorithm to eliminate false match points.In this paper,the CCTV network codes as experimental subjects,comparative analysis of the experimental results,to improve the accuracy of matching character verification code.