Assessing motion-coherency in video watermarking
P. Vinod, Gwenaël Doërr, Prabin Kumar Bora · 2006
Motion coherent watermarking has been recently proposed as a means to combat temporal frame averaging along the motion axis (MC-TFA). The fundamental idea consists in exploiting motion-compensation primitives to force a physcal point of the scene to always carry the same watermark sample wherever it is projected in the video. However, for a given watermarking system, there is no simple tool to assess whether the produced watermark is motion-coherent or not. Today, this assessment relies on a computationally expensive procedure, namely (i) embed a watermark, (ii) perform the MC-TFA attack, (iii) check for the presence of the watermark. Therefore, the goal of this article is to provide the community with an efficient and accurate oracle which reports whether a video sequence contains any non-motion coherent component or not. This is done in practice by looking at the statistics of the difference between a frame and a motion predicted version of it.