Spd Attack
AI image editing tools are becoming incredibly powerful and easy to use. With just a few clicks, these tools can transform images in ways that used to take hours of expert work. While this technology opens up amazing creative possibilities and makes advanced editing accessible to everyone, it also creates new risks. Images can be manipulated to create misleading content, fake identities, or violate artists’ rights.
This blog introduces Safeguarded Perturbation Defense (SPD) Attack, a new way to “immunize” images against AI manipulation. Our method builds on recent works that immunize images against adversarial attackers. Particularly, Immunization is achieved by adding subtle, imperceptible perturbations into images, reinforcing their resistance to unintended image editing. These perturbations interfere with the functionality of generative models, inhibiting image manipulation while preserving the original visual characteristics and perceptual fidelity.
Current methods generate a single adversarial noise to a single image, in our blog we explore how to extend these approaches for multiple images.