Abstract

We present schemes for hardening image watermarks against adversarial jamming. Our techniques improve upon standard correlation- and spread spectrum-based methods, and withstand various image distortions and attacks intended to render watermarks unreadable. The watermarking schemes we propose explicitly locate and amplify watermark data remaining after attacks. Key ideas include embedding of watermarks in specially chosen domains, application of image enhancement, computation of responses overwatermark subsets, and use of redundancy and search. For watermark detection, our schemes do not require the original image or any information about it.

‚Äč