Robust image watermarking using local Zernike moments

Abstract

In this work, we propose a robust image watermarking algorithm using local Zernike moments, which are computed over circular patches around feature points. The proposed algorithm locally computes Zernike moments and modifies them to embed watermarks, achieving robustness against cropping and local geometric attacks. Moreover, to deal with scaling attacks, the proposed algorithm extracts salient region parameters, which consist of an invariant centroid and a salient scale, and transmits them to the decoder. The parameters are used at the decoder to normalize a suspect image and detect watermarks. Extensive simulation results show that the proposed algorithm detects watermarks with low error rates, even if watermarked images are distorted by various geometric attacks as well as signal processing attacks.

Introduction

Digital images can be easily duplicated without any loss. This feature facilitates the illegal use of copyrighted materials, e.g., unrestricted duplication and dissemination via the Internet. As a result, the piracy makes service providers hesitate to offer services in digital form, in spite of digital imaging equipments replacing analog ones. To overcome this reluctancy and possible copyright issues, the intellectual property rights of digital images should be protected.

To protect copyrighted images, many approaches, including authentication, encryption, and digital watermarking, have been proposed. Encryption methods may guarantee secure transmission of data to authenticated users through unsecured channels. However, once decrypted, the data are identical to the original and their piracy cannot be restricted. Digital watermarking is an alternative approach to deal with these unlawful acts. It hides invisible marks or copyright information in digital contents and claims the copyrights. The marks should be robust enough to survive various attacks. It is also desirable that illegal attempts should cause the degradation of image quality without erasing the watermarks.

With the development of watermarking technologies, attacks against watermarking systems also have become more sophisticated. Those attacks can be classified into signal processing attacks and geometric attacks. Signal processing operations, such as lossy compression, denoising, noise addition, and filtering, reduce the energy of watermarks, while geometric attacks induce synchronization errors between the original and the watermarked images and therefore can mislead the watermark decoder. Most previous watermarking methods provide robustness against signal processing attacks, but only a few specialized methods address geometric attacks. Watermarking schemes to combat geometric attacks can be classified into non-blind schemes, semi-blind schemes, invariant domain embedding, template-based synchronization, and content-based synchronization. Section 2 will review these watermarking schemes.

In this paper, we propose a novel robust image watermarking algorithm. Although there are techniques that deal with geometric attacks, random bending, cropping, and non-isotropic scaling still remain to be difficult attacks. The proposed algorithm embeds watermarks by modifying local Zernike moments (LZMs), which are defined over circular patches around feature points. Unlike the conventional algorithms using moments [1], [2], we compute Zernike moments locally to achieve resilience against cropping. Moreover, to make the proposed algorithm robust against scaling attacks, we extract salient region parameters and transmit them to the decoder. The decoder uses the parameters to normalize a suspect image and detect watermarks. Simulation results demonstrate that the proposed algorithm provides robust performance in various attack scenarios. We presented some of our preliminary results in [3]. In this work, we extend the work in [3] by evaluating the performance of the proposed algorithm more extensively in multiple application scenarios. Moreover, we analyze the false alarm rate of the proposed watermarking algorithm.

The paper is organized as follows. Section 2 reviews conventional synchronization techniques. Section 3 introduces local Zernike moments and a feature point detection scheme. Section 4 describes the proposed watermarking system. Section 5 proposes an image normalization scheme using salient region parameters. Section 6 evaluates the performance of the proposed algorithm. Section 7 concludes the paper.