Image watermarking with better resilience

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 the embedding of watermarks in specially chosen domains, the application of image enhancement, the computation of responses over watermark subsets, and the use of redundancy and search. For watermark detection, our schemes do not require the original image or any information about it.