Recovering Invisible Image Watermarks from Images Distorted by the "StirMark" Algorithm

Abstract

"StirMark" is an image distorting algorithm developed at Cambridge University that is intended to attack and obliterate robust invisible image watermarks. The algorithm modifies a watermarked image by geometric distortion so subtle that the modification is essentially unnoticeable to a human observer. However, its effect on imbedded robust invisible watermarks can be devastating, successfully rendering them undetectable. In this paper, a countermeasure to the StirMark algorithm will be presented. It involves first detecting the presence of distortion in the distorted image, then measuring the magnitude and type of distortion, and finally removing the measured distortion thus restoring image geometry. Using a robust invisible watermarking method presented previously by the author, the results of removal of StirMark distortions will be demonstrated by showing successful extraction of an imbedded watermark from a realigned image.