Gaussian Hypothesis for Video Watermarking Attacks: Drawbacks and Limitations
Corneliu Octavian Dumitru, Sorin Duta, Mihai Mitrea, Francoise J. Preteux · 2007
In the absence of any theoretical or experimental support, the video watermarking attacks are by default assumed to be Gaussian distributed. By combining four types of statistical tests, the present paper carries out an in-depth study on this hypothesis for eight types of attacks: filtering (median, Gaussian, FMLR, sharpening), rotations, JPEG compression, row&column removal, and StirMark. These attacks are investigated in the hierarchy of the DCT and DWT coefficients. The results in this paper show that the Gaussian assumption can be rejected for most of the attacks. Finally, the non-Gaussian behaviour impact on several applications connected to video watermarking capacity is discussed.