Exploration of 3D Texture and Projection for New CAPTCHA Design

Abstract

Most of current text-based CAPTCHAs have been shown to be easily breakable. In this work, we present two novel 3D CAPTCHA designs, which are more secure than current 2D text CAPTCHAs, against automated attacks. Our approach is to display CAPTCHA characters onto 3D objects to improve security. We exploit difficulty for machines in rotating 3D objects to find a correct view point and in further recognizing characters in 3D, both tasks that humans can easily perform. Using an offline automated computer vision attack, we found that 82% of the new text reCAPTCHA characters were successfully detected, while approximately 60% of our 3D CAPTCHAs were detected only if characters were focused and zoomed from the direct view point. When CAPTCHAs are presented in slightly different views, the attack success rates against our approaches are reduced to almost 0%.