Comparison of perceptual shaping techniques for digital image watermarking

Digital watermarking is the process of embedding information into another signal like an image. Embedding a watermark results in modifications of the image, leading to a decrease of the perceived quality. Perceptual shaping uses perceptual information of the image to camouflage these modifications, thereby increasing the perceived quality. In this paper four perceptual models are evaluated: the empirical noise visibility function, the noise visibility function based on a stationary generalized Gaussian distribution, the noise tolerance model and the Watson model. These models are applied on two watermarking schemes: correlation-based watermarking and dither-modulated quantization index modulation. We describe how these schemes utilize the perceptual models for improving the perceived quality and experimentally compare their performance. We conclude that the noise tolerance model offers the best results regarding the perceived quality. The empirical noise visibility function was found to offer best robustness.