Chinese Style Matching CAPTCHA: Style Matching Technology Based on Deep Learning
Xuan Song, Jialiang Zhu, Liping Du · 2025
With the rapid development of the internet, CAPTCHAs (Completely Automated Public Turing tests to tell Computers and Humans Apart) have become a critical line of defense against malicious automated attacks. However, the security of traditional text-based and image-based CAPTCHAs is facing severe challenges from deep learning technologies. This study innovatively proposes a Chinese Style Matching CAPTCHA (CSMC) based on Neural Style Transfer (NST), aiming to significantly enhance CAPTCHA security while maintaining excellent user experience. This research comprehensively explores the developmental history of CAPTCHA technologies, analyzes the vulnerabilities of traditional CAPTCHAs in the context of deep learning, and focuses on the current applications of neural style transfer technology in the CAPTCHA domain. Building on this foundation, we meticulously design a novel Chinese Style Matching CAPTCHA system (CSMC) that ingeniously integrates neural style transfer technology from deep learning. By generating stylized images and requiring users to match the styles used to create these images, the system substantially increases the difficulty for automated programs to recognize them. Furthermore, this study rigorously evaluates the proposed method's performance in resisting automated attacks and security analysis. The results demonstrate that the CSMC system achieves a balance between security and user experience while providing innovative approaches for CAPTCHA design.