Robust Image Watermarking Based on Scale-Space Feature Points

Abstract

Digital watermarking techniques have been explored extensively since its first appearance in the 1990s. However, watermark robustness to geometric attacks is still an open problem. The past decade has witnessed a significant improvement in the understanding of geometric attacks and how watermarks can survive such attacks. In this chapter, we will introduce a set of image watermarking schemes which can resist both geometric attacks and traditional signal processing attacks simultaneously. These schemes follow a uniform framework, which is based on the detection of scale-space feature points. We call it the scale-space feature point based watermarking, SSFW for short. Scale-space feature points have been developed recently for pattern recognition applications. This kind of feature points are commonly invariant to rotation, scaling and translation (RST), therefore they naturally fit into the framework of geometrically robust image watermarking. Scale-space feature points are typically detected from the scale space of the image. As a result, we will first introduce the scale space theory and how the feature points can be extracted. The basic principles on how the scale-space feature points can be adapted for watermark synchronization are then discussed in detail. Subsequently, we will present several content-based watermark embedding and extraction methods which can be directly implemented based on the synchronization scheme. A detailed watermarking scheme which combines scale-invariant feature transform (SIFT) and Zernike moments is then presented for further understanding of SSFW. Watermarking schemes based on the SSFW framework have the following advantages: (a) Good invisibility. The Peak Signal to Noise Ratio (PSNR) value is typically higher than 40dB. (b) Good robustness. These schemes can resist both signal processing attacks and geometric attacks, such as JPEG compression, image filtering, added noise, RST attacks, locally cropping as well as some combined attacks.