Source camera identification for securing AI-enabled surveillance cameras

The use of IP-based smart cameras for video surveillance has increased tremendously over the last decade with an estimate of over 1 billion surveillance cameras installed worldwide. The recorded images and videos are used by professionals for crime prevention, traffic management, and other data driven analysis. By launching different types of attacks on the IP cameras, the source images and video footage can be corrupted, which would lead to wrong decision making. In this paper, we have proposed to use source camera identification using a tunable filter bank based feature in a video surveillance system for authenticating the original source of an image or video taken from a smart IP camera. For the feature matching several distance measures are used. With Canberra distance, we have achieved state-of-the-art accuracy using 36 features.