A method for hiding synchronization marks in scale and rotation resilient watermarking schemes

As concern about watermark robustness growns, different approaches based on the introduction of synchronization marks have been presented to survive geometrical attacks. Most of the proposed techniques rely either on the introduction of a dedicated template either on the detection of particular properties in the autocorrelation function (ACF) of the watermark itself in order to resist against scaling and rotation transformations. The use of this side-information at the detector makes an inversion of the transformation possible. However, due to the publicness of this side-information, those techniques turn out to be very vulnerable to a removal attack. We propose an innovative method to hide the synchronization marks and therefore prevent malicious removal attacks (eg. template attack,...). The ability to detect the synchronization marks will be conditionned by the knowledge of a secret key. The technique consists in using an image dependent secret binary mask to modulate the synchronization pattern before it is introduced in the image. The ability to recover this binary mask upon rotation and scaling allows the detection of the synchronization marks even after transformation. Although mask recovering presents a considerable error rate, sufficient detection of the synchronization marks can be achieved. The mask, obtained from a signal-dependent partition, leads to a spectral spreading of the synchronisation mark, making template attacks nearly impossible to perform.