Robust and reliable image copyright protection scheme using downsampling and block transform in integer wavelet domain

Highlights

• Embed each sub-watermark obtained by downsampling into corresponding sub-host image.

• Encrypt four sub-watermarks by pixel scrambling and DFRNT to enhance its security.

• Embed the right principal component of watermark into the host image to avoid FPP.

• Apply two-level SVD to host image and achieve the double embedding of watermark.

• Optimize the embedding factors by GDPSO to balance the invisibility and robustness.

Abstract

In this paper, a robust watermarking algorithm in integer wavelet domain using downsampling is proposed. The innovations of this paper can be summarized as follows. First, after downsampling both the host image and watermark, each sub-watermark is embedded into the corresponding sub-host image. Second, four sub-watermarks are encrypted by pixel scrambling and fractional random transform, and the right principal component of watermark is embedded in the host image to avoid false positive problem. Third, two-level singular value decomposition and block cosine transform are performed on host image and watermark respectively, and the dual embedding of watermark is realized. Fourth, guided dynamic particle swarm optimization is used to optimize the embedding factors. The simulation results show that the proposed watermarking algorithm satisfies the requirements of robust watermarking very well. It has large capacity and strong robustness to common attacks. Moreover, compared with the existing related algorithms, this algorithm has obvious advantages.

Introduction

In the era of rapid development of multimedia information technology, people can obtain needed resources in a variety of ways. The data can be easily and completely duplicated, which brings great convenience to human life, work, scientific research and so on [1]. However, the information security issues have arisen, which has attracted great attention. Copyright ownership is an important aspect of information security, and digital watermarking is an effective way to achieve copyright protection [2]. It achieves the purpose of copyright protection by embedding the content representing copyright information into the digital works to be protected. In recent years, many watermarking algorithms have been proposed [1], [2], [3], [4], [5], [6], [7], [8], [9], [10], [11], [12], [13], [14], [15].

There are many classification methods of watermarking algorithm. According to characteristics, it can be divided into robust watermarking and fragile watermarking [16], [17]. The former is mainly used in the copyright protection of digital works, which requires strong robustness and security. The latter is often used for integrity protection and authentication of digital works [18]. When the content of digital works changes, the watermark information will change accordingly, which can identify whether the original data has been tampered with [19]. Most of the current research focuses on robust watermarking algorithm for copyright protection, such algorithm usually embeds the watermark into the transform domain of the host image to enhance robustness. Discrete Fourier transform (DFT), discrete wavelet transform (DWT) and discrete cosine transform (DCT) are the most commonly used transform domain methods [9], [10], [14], [20], [21], [22], [23], [24], [25], [26], [27]. In addition, some time-frequency analysis tools, such as fractional wavelet transform (FWT) [6], fractional Fourier transform (FRFT) [28], linear canonical transform (LCT) [29], Contourlet transform (CT) [30], etc., are also used in information security, especially in digital watermarking [6], [16], [31], [32], [33], [34], [35]. Because the robust watermarking algorithm also has high requirement for security, different encryption schemes have been proposed and used in watermarking algorithm to enhance the security of the algorithm. In image encryption, besides the widely used chaotic mapping [8], [36], there are time-frequency transform methods such as gyrator transform (GT), FRFT and LCT [36], [37]. This paper mainly studies the robust watermarking. For the requirements of such watermarking, a watermarking algorithm with high security and robustness needs to be designed.

DWT has important applications in the copyright protection of digital works. Its variants, integer wavelet transform (IWT) and redundant wavelet transform (RWT), can also be used for digital watermarking. Many related algorithms have been proposed [10], [19], [20], [22], [38], [39], [40], [41], [42], [43], [44]. In addition, singular value decomposition (SVD) has excellent properties and is often used to improve the robustness of watermarking algorithm. However, because of the high computational complexity, it is often combined with transform domain method to design watermarking algorithm [1], [8], [22], [24], [38], [39], [45], [46]. Makbol et al. [38] proposed to apply IWT to host image, then add the watermark directly to the singular values of all sub-bands in host image. Moreover, a signature scheme was proposed to avoid false positive problem (FPP). Unlike [38], Ansari et al. [39] only selected the low and middle frequency sub-bands of the host image to hide the watermark, in which the embedding factor was optimized by artificial bee colony algorithm. Furthermore, this algorithm gave a new signature scheme. In the above two algorithms [38], [39], the signature scheme has been presented to solve the FPP. However, on the one hand, the FPP cannot be completely avoided in this way. On the other hand, the generation, embedding and extraction of signatures make the watermarking algorithm complicated, increasing the computational cost of the algorithm. In addition, the security of these two algorithms is low, and they also shows poor robustness in resisting scaling attack. In the algorithm proposed by Makbol et al. [43], a new method for solving FPP was introduced, in which the left singular vector of the watermark was embedded into the singular value of the integer wavelet domain in the host image. Although this algorithm completely solves FPP, its security is low. Moreover, since the SVD is directly applied to the watermark, the time complexity of this algorithm is high. Furthermore, the robustness of this algorithm against scaling, translation and sharpen attacks is weak. Zhou et al. [20] selected the high frequency of the host image wavelet domain and applied $8\times 8$ block DCT to it, then extracted the medium frequency coefficients of each block to construct a new matrix and performed fractional random transform (FRNT) on it. Finally, the encrypted watermark was added to the FRNT coefficients of host image. This algorithm has a small watermark capacity. In addition, when watermarked image is subjected to JPEG compression, scaling, translation and cropping attacks, the quality of extracted watermark is very poor. In the algorithm presented by Hurrah et al. [19], the low-frequency of host image in the wavelet domain was divided into $8\times 8$ blocks, and each block was further divided into four $4\times 4$ blocks, then each sub-block was processed by DCT. Finally, the encrypted watermark bits were embedded by modifying the difference between the selected two DCT coefficients. Compared with [20], the watermark capacity of this algorithm is lower. Besides, the transparency of the algorithm is low, and the resistance of this algorithm to gaussian noise, scaling and translation attacks is weak. Arunkumar et al. [45] first applied $8\times 8$ block redundant integer wavelet transform to the host image, then performed DCT to the low frequency of each block and encrypted watermark. After dividing the DCT coefficient of the watermark into $4\times 4$ blocks, the singular value of each watermark block is added to the corresponding host image block. The transparency and security of this algorithm are high, but its robustness is poor, especially for Gaussian noise, speckle noise, salt & pepper noise, scaling, translation and sharpening attacks. To sum up, the robustness of these algorithms against some common attacks is weak, and the capacity is small. Moreover, the security needs to be improved.

In view of the problems mentioned above, we propose a secure and robust watermarking algorithm by introducing downsampling, two-level SVD and block DCT. Firstly, the host image and watermark are preprocessed by downsampling, and four sub-watermarks are double-encrypted with pixel scrambling and discrete fractional random transform to enhance its security. The computation speed of block DCT and SVD is much faster than global transform. Therefore, two-level SVD is used to obtain the corresponding singular value for the low and high frequency parts of the integer wavelet domain in the host image. The low-frequency part of watermark image in integer wavelet domain is transformed into DCT domain by block DCT, then its singular value is obtained with SVD. Finally, the right principal component of each sub-watermark is embedded into the singular value matrices of corresponding sub-host image in the low and high frequency sub-bands. In order to balance the invisibility and robustness, we use guided dynamic particle swarm optimization to obtain the optimal embedding factors. The double embedding of the watermark improves the robustness of the watermarking algorithm. Based on the experimental results, the proposed algorithm has good transparency and security, and the watermark with the same size as the original image can be hidden in the digital image. Moreover, this algorithm is robust to common attacks, especially for JPEG compression, scaling, translation and various filtering attacks.

The rest of this paper is organized as follows. Some theories used in this algorithm are introduced in Section 2. Section 3 gives a detailed description of proposed algorithm, including the embedding and extraction processes, as well as the optimization of the embedding factors. Simulation results and the comparison with existing related algorithms are presented in Section 4. Finally, the conclusion is given in Section 5.