Security Analysis of ISS Watermarking Using Natural Scene Statistics

Watermarking security has captured great attention from researchers in recent years. The security of watermarking is determined by the difficulty of estimating the secret key used in embedding/detecting schemes. As a widely used scheme, Improved Spread-Spectrum (ISS) watermarking performs better than Additive Spread-Spectrum (Add-SS) and is able to include Add-SS watermarking as a specialized case. Because of its popularity, the investigation on the security of ISS watermarking has been reported. Previous works on evaluating the security of ISS watermarking mainly focus on the assumption of Gaussian host and ignore the effects from the non-Gaussian characteristics of natural images. This paper analyzes the security of ISS watermarking from the viewpoint of Shannon information theory by using Gaussian Scale Mixture (GSM) model to characterize the natural scene statistics and reveals the relationship between the security and its related factors. Theoretical analysis and simulation results show that the security of ISS watermarking with the Gaussian host assumption is over-stated in previous work.