Defensive Patch

Defensive patches are small, strategically designed image modifications used to protect computer vision systems from adversarial attacks, such as those aiming to fool object detectors or OCR systems. Current research focuses on developing robust and efficient patch designs, exploring both active defense strategies (e.g., injecting counter-patches) and certified defenses that guarantee robustness. These methods aim to improve the resilience of computer vision models against real-world attacks with minimal impact on model accuracy and computational cost, impacting applications ranging from autonomous driving to security systems.