Saliency-Aware Privacy Protection in Augmented Reality Systems

Abstract

The augmented reality (AR) Metaverse environment combines the physical and virtual world together. Privacy is a major concern in AR since the cameras use to capture the physical world can also capture other images that may potentially violate user or by-stander privacy. Advances in deep learning to process images and videos have exacerbated such privacy risks. This paper presents a new technique to protect privacy in AR systems by combining the idea of visual saliency together with privacy-sensitive object detection. We show that our technique is able to provide additional context to a given image to better balance between privacy and overall usability of the system.