Results of attacks on a claimed robust digital image watermark

Abstract

Placing either a visible or invisible mark on a document to help establish its ownership is not a new or revolutionary concept. For hundreds of years, owners of important documents or works of art have imprinted identifying marks onto them, not only to help establish their ownership, but also to discourage those who might wish to misappropriate the work. Today, invisible watermarking of digital images is attempted for those same purposes. To be useful, an ownership watermark must robustly survive and be detectable after may manipulation that does not damage a digital image beyond usability. In the extreme, the watermark must survive the printing and rescanning of the image. Additionally, the probability of detecting a watermark in an image that does not have one must be vanishingly small. An embodiment of invisible watermarking claiming to have these properties has been presented previously by the author. This paper reports the results of typical image manipulations and deliberate attacks on robust invisible watermarks of the type reported. Image manipulations include lossy JPEG compression, small angle rotation, linear and nonlinear resizing, cropping, and sharpening. Attacks include the superposition of uncorrelated noise fields, overmarking, and RSS alterations.