A Cauchy distribution-based ternary hypothesis testing for bipolar additive watermark detection in H.264/AVC video

Abstract

The statistical distribution of discrete cosine transform (DCT) coefficients is important for video watermarking since they are the main carriers for watermark embedding. For the statistical distribution of intra-coded DCT coefficients in the H.264/AVC video stream, non-parametric hypothesis test is utilised to verify that Cauchy distribution is better than generalised Gaussian distribution (GGD). Moreover, ternary hypothesis test is introduced into the detection of bipolar additive watermarks. By adjusting the watermark strength parameter, the detector performance can be guaranteed. Experimental results show that for those bipolar additive watermarks in the H.264/AVC stream, the proposed approach can achieve a detection accuracy of more than 80% on average.

Keywords:

• H.264/AVC
• digital watermarking
• Cauchy distribution
• ternary hypothesis testing

This work is supported in part by the Special Pro-phase Project on National Basic Research Program of China (2010CB334706), National Natural Science Foundation of China (61072122), Key Project of Hunan Provincial Natural Science Foundation (11JJ2053) and the Fundamental Research Funds for the Central Universities (Hunan University).