A Review on Text-based CAPTCHA Breaking Based on Deep Learning Methods

摘要

The security of cyberspace has become a global common concern, with websites and information systems facing significant threats. Text-based CAPTCHAs, widely implemented by websites, play a crucial role in preventing illicit attacks by robots and crawlers. However, in recent years, various breaking methods, particularly those based on deep learning, have emerged. This paper conducts a comprehensive investigation into the resistance mechanisms of text-based CAPTCHAs, analyzing their technical aspects in CAPTCHA design. Subsequently, we explore the recognition procedures of CAPTCHA utilizing deep learning models and compare representative algorithms developed in recent years. Finally, we evaluate the positives and negatives of different cracking algorithms and summarize our findings.