Enhancing Image-Based Captcha Scheme Using Cycle-Consistent Generative Adversarial Network

Abstract

Researchers have carried out extensive work on improving Completely Automated Public Turing Test to Tell between Computers and Humans Apart (CAPTCHA) to prevent malware or bots from compromising information, this has led to the development of the most secure and user-friendly form of CAPTCHA popularly known as image-based CAPTCHA. It is best used as a shield that protects unauthorized access to information available online. As more sophisticated algorithms emerge, attacks on image-based CAPTCHA have also increased, and using deep learning algorithms attacks on CAPTCHA design have become more vulnerable. Research has proven that Adversarial has a promising direction in overcoming such challenges, hence we proposed an image-based CAPTCHA scheme as an effort to enhance the CAPTCHA design through the use of cycle-consistent generative adversarial network which was minimized with Mean Square Error and Mean Absolute Error and Inception Score was used to evaluate the quality of generated image with an average difference of 0.025 when compared with the existing scheme. That is to say our scheme effectively produces a synthetic image that is indistinguishable from the real image, which can easily fool the DeCAPTCHAs while solving an automated CAPTCHA challenge.