On the Security and Applicability of Fragile Camera Fingerprints
This addresses the problem of image forensics and authentication security for digital camera users, providing an incremental analysis of an existing defense mechanism.
The paper tackles the security of fragile camera fingerprints against adversarial attacks, demonstrating through theoretical and practical tests that they enable reliable device identification under common compression levels.
Camera sensor noise is one of the most reliable device characteristics in digital image forensics, enabling the unique linkage of images to digital cameras. This so-called camera fingerprint gives rise to different applications, such as image forensics and authentication. However, if images are publicly available, an adversary can estimate the fingerprint from her victim and plant it into spurious images. The concept of fragile camera fingerprints addresses this attack by exploiting asymmetries in data access: While the camera owner will always have access to a full fingerprint from uncompressed images, the adversary has typically access to compressed images and thus only to a truncated fingerprint. The security of this defense, however, has not been systematically explored yet. This paper provides the first comprehensive analysis of fragile camera fingerprints under attack. A series of theoretical and practical tests demonstrate that fragile camera fingerprints allow a reliable device identification for common compression levels in an adversarial environment.