BCH coding and intelligent watermark embedding: Employing both frequency and strength selection

Abstract

This paper presents a novel approach of adaptive visual tuning of a watermark in Discrete Cosine Transform (DCT) domain. The proposed approach intelligently selects appropriate frequency bands as well as optimal strength of alteration. Genetic Programming (GP) is applied to structure the watermark by exploiting both the characteristics of human visual system and information pertaining to a cascade of conceivable attacks. The developed visual tuning expressions are dependent on frequency and luminance sensitivities, and contrast masking. To further enhance robustness, spread spectrum based watermarking and Bose–Chadhuri–Hocquenghem (BCH) coding is employed. The combination of spread spectrum sequence, BCH coding and GP based non-linear structuring makes it extremely difficult for an attacker to gain information about the secret knowledge of the watermarking system. Experimental results show the superiority of the proposed approach against the existing approaches. Especially, the margin of improvement in robustness will be of high importance in medical and context aware related applications of watermarking.

Introduction

To implement Digital Rights Management (DRM), digital watermarking has received substantial attraction. A very recent example is the focus of Distributed Computing Industry Association on “how digital watermarks can ensure DRM in peer-to-peer Applications”. As regards DRM, two important but contradicting properties of a watermarking system; robustness and imperceptibility, are of prime importance. Making a tradeoff between these two properties in view of both human visual system (HVS) and the anticipated application is a challenging problem. The intended application of the watermarking scheme needs a specific balance between these properties. Therefore, treating this balance achievement as an optimization problem and developing strategies that may adaptively deliver this delicate balance for a new application has attracted many researchers.

A watermark is generally structured using perceptual models, which exploit the characteristics of HVS in order to attain imperceptibility. These perceptual models are able to learn the frequency content of a cover image by exploiting the sensitivities/insensitivities of HVS. They take advantage of frequency sensitivity models that are based on viewing conditions as well as the cover image dependent, luminance sensitivity and contrast masking effects. Frequency sensitivity describes the HVS sensitivity to sine wave gratings at different spatial frequencies and depends only on the surrounding conditions. Luminance sensitivity on the other hand, is a measure of the effect of the detectable threshold of a signal on a constant background. It depends on the average luminance value of the background as well as on the signal's luminance level. In block-based Discrete Cosine Transform (DCT) case, the DC coefficient of each block dictates the luminance sensitivity for that block. The third important property of HVS that is exploited for hiding a watermark is the contrast masking. It represents the ability of detecting one signal in the presence of another signal. This masking effect increases when the masking signal and the signal to be masked have same spatial frequency, orientation and location. In block-based DCT, the AC coefficients dictate this behavior.

Watson's Perceptual Model [1] is an appropriate example of such models, as it has been extensively used in image compression and digital watermarking [2], [3]. Other pertinent examples of such models include Lambrecht's Gabor filters based model [4], and Kutter's model based on local isotropic measure and a masking function [5]. However, all these perceptual models are devised in view of invisibility requirements alone, and thus, lack in exploiting the information pertaining to distortions caused by a set of probable attacks. As a result, they are inefficient to cater, concurrently, for both robustness and imperceptibility needs of a watermarking system. These approaches are thus inappropriate for applications, such as online catalogs, medical imagery and Print-to-Web technology, which, besides imperceptibility, require a high degree of robustness as well. In [3], an effort is put to cope with this problem; however, their approach does not perform frequency band selection and thus is not able to survive a series of attacks on the watermarking system.

A variety of watermarking schemes are proposed in literature, including transform domain techniques [6], [7], [8], [9], and spatial domain methods [10]. Almost all of the transform domain watermarking approaches make use of a predetermined set of coefficients, mostly in the middle frequency range, to serve as a tradeoff for watermark embedding. The disadvantages of such a selection are discussed in detail by Shieh et al. [11], who also proposed a scheme in which suitable embedding positions in a block based DCT domain watermarking are selected using Genetic Algorithm. Nevertheless, they do not take into consideration the allowable strength of alteration and robustness against a series of attacks. Harrak et al. [12] proposed a GP based watermarking technique which makes a tradeoff in terms of image quality and robustness. The drawbacks of their scheme include non-blind extraction of the watermark, no position selection and inability to preserve the watermark after even a single unintentional attack like Gaussian noise or JPEG compression. Lee et al. [13] proposed a hybrid watermarking scheme based on Genetic Algorithm and Particle Swarm Optimization in Discrete Wavelet Transform (DWT) domain. Their work is based on the assumption that higher detection response could be expected if two complementary watermarks are simultaneously embedded into the transformed coefficients through positive and negative modulations, respectively. Nevertheless, each type of modulation resists a specific class of attacks, and thus, limits the watermark detection in case where different classes of attacks are presented in a sequence. Lu et al. [14] have proposed a robust watermarking scheme based on DWT and non-negative matrix factorization, which is robust against a single attack. Li and Wang [15] have achieved the robustness–imperceptibility tradeoff against individual attacks by combining Radon transform with 2D wavelet transform. Chang et al. [16] used Singular Value Decomposition (SVD) in an attempt to achieve a subtle balance between the robustness and imperceptibility requirements. On the other hand, Zheng et al. [17] have used Particle Swarm Optimization (PSO) in integer DCT domain to optimize the two watermarking objectives. Recently, Aslantas et al. [18] have proposed the idea of utilizing SVD and PSO in a DWT based image watermarking technique. Their proposed idea is to intelligently select the optimum scaling factors for different wavelet sub-bands in view of different attacks. The drawback of their scheme is that it does not allow the blind extraction of the watermark and is robust against a single attack only. Khan [19] also presented a GP based attack resistant watermarking approach.

Therefore, the performance of the existing watermarking approaches is not up to the task when we consider watermark structuring in view of a sequence of attacks, which is much desirous in real world applications. In the present work, in order to resist a series of attacks, we employ selection of both frequency band, as well as, strength of alteration for watermark embedding simultaneously.

Error correcting codes (ECCs) are gaining more popularity in terms of their capability to enhance watermark robustness. Alattar et al. have used spread spectrum technique and BCH coding for watermarking electronic documents [20]. The work by Miaou et al. [21] demonstrates significant improvement in robustness by using BCH coded watermarks in error-prone transmission of MPEG video. Other pertinent examples include the use of low-density parity check codes (LDPC) [22], and Reed-Solomon codes [23] as error correcting codes to enhance robustness in digital watermarking.

The remaining part of this paper is organized as follows: Section 2 introduces perceptual modeling. Our proposed method is described in Section 3, which includes GP simulation, implementation of frequency band selection through the concept of a wrapper function and application of BCH codes in the proposed watermark structuring approach. Implementation details are presented in Section 4. Section 5 discusses the potential applications of the proposed approach. Results and discussion are presented in Section 6, followed by conclusion in the last section.