Font Size: a A A

Verification Code Decoding Algorithm Research And Implementation

Posted on:2014-02-10Degree:MasterType:Thesis
Country:ChinaCandidate:S F YangFull Text:PDF
GTID:2248330395482519Subject:Control theory and control engineering
Abstract/Summary:PDF Full Text Request
The development of the Internet has provided people with a life of convenience. Free network services gradually become a part of people’s lives, such as E-mail, internet chat system and electronic forums. However, unauthorized program attacks at these resources, which also brings new challenges for internet security problem. CAPTCHA appear for this reason. Research CAPTCHA breaking algorithm can improve the defects of CAPTCHA. It has important significance in protective unauthorized program’s attack and can enhance internet security. The main work of this paper is as follows:(1) Several digital image processing algorithms applied to pre-processing of CAPTCHA is designed. A threshold processing algorithm based on HSL color space is designed because of one type CAPTCHA processing failed by iterative method and OTSU algorithm. The experiments show that the effectiveness of this algorithm.(2) In the character recognition stage, firstly, neural network theory is introduced. Secondly, character recognition algorithm based on a simplified convolutional neural network is implmented. Finally, elastic deformation method is implemented for improving the performance of character recognition system.(3) Characters is non-touching or touching after CAPTCHA pretreatment. According to this feature, non-touching characters extraction algorithms is implemented, which are projection method and region extraction algorithm. Then, we focus on touching characters CAPTCHA segmentation algorithm. Firstly, segmentation algorithm based on the largest connected region is proposed. Next, the segmentation algorithm based on background-thinning is implemented and improved.(4) As the typial exmples of non-touching characters CAPTCHA projection method is used to break ICBC’s CAPTCHA and region extraction algorithm is applied for58city’s CAPTCHA. As the typial exmples of touching characters CAPTCHA improved segmentation algorithm based on background-thinning is used to break Hotmail’s CAPTCHA and Tencent’s CAPTCHA. Breaking rate of those CAPTCHAs are94.5%,41%,61%and42%, respectively.(5) Three tools software modules for accelerating CAPTCHAs breaking research is designed.
Keywords/Search Tags:CAPTCHA breaking, Convolutional neural network, Characters segmentation, Character recognition
PDF Full Text Request
Related items