Today, vast amounts of visual information are collected and often shared online. Such images and videos can contain various types of privacy-sensitive information that can be extracted automatically at a large scale, posing a steadily growing threat to user privacy. We address core scientific challenges by understanding and enforcing privacy in visual information as well as working towards our overall vision of a Visual Privacy Advisor.
Understanding Privacy Risks in Visual Information. We investigate what type of private information can be contained in visual information. With the rapid development of advanced machine learning techniques – in particular Deep Learning – rich and high-level semantic information can be extracted from the raw image data that can include identities, social relations, habits and behaviors. Understanding the quality and quantity of this information is key to assess risks and develop counter measures.
Enforcing Privacy in Visual Information. We research foundations and methods to enforce privacy policies. As there is natural tension between communication, utility and privacy, working out compromises is of key importance. In particular, advanced machine learning techniques are investigated to automatically sanitize, edit and resynthesize multi-modal content in order to protect privacy.
Towards a Visual Privacy Advisor. We work towards the overall vision of a Visual Privacy Advisor that assists the user in assessing and mitigating privacy risks in visual data. This line of research connects the understanding of privacy risks and methods of enforcing privacy in order to help users navigating and acting in this complex space. The methodological advances are coupled with a user perspective that brings about personalization and long term privacy objectives.