Geometry-Based CAPTCHA Design for Bot Resistance Using Shape Detection

This study proposes a geometry-based CAPTCHA system designed to improve resilience against automated attacks by leveraging shape detection and human perceptual principles. The system dynamically generates challenge images composed of randomly placed triangles and rectangles embedded in varying levels of visual noise, eliminating the need for static image databases. Users are required to identify and count specific shapes, a task informed by Gestalt principles that remain difficult for machine learning models to replicate. To evaluate robustness, the YOLOv5 object detection model was employed to simulate bot-based attacks, and the system was benchmarked against reCAPTCHA, ShapeCAPTCHA, and Puzzle CAPTCHA. Usability was assessed through a controlled user study. Experimental results demonstrate that the proposed approach offers improved resistance to automated solvers while maintaining a high level of usability, making it a scalable and lightweight alternative to conventional CAPTCHA systems.