Elsevier

Signal Processing

Volume 90, Issue 8, August 2010, Pages 2498-2512
Signal Processing

A robust blind color image watermarking based on wavelet-tree bit host difference selection

https://doi.org/10.1016/j.sigpro.2010.02.017Get rights and content

Abstract

In this paper, a blind watermarking technique based on the so called wavelet-trees is introduced. The proposed technique deals with the color pixel as one unit and exploits the significant features and relations between the color pixel components in the wavelet domain. The watermark is embedded by spreading it through the host image in such a manner that the inter-pixel robust relations carry the watermark bit sign with sufficient energy. Experimental results have shown that the proposed scheme is highly imperceptible with PSNR=41.78−48.65 dB, for various watermarked images, with a capacity of 3072 bits. Also, the proposed scheme is highly resistive to various common signal processing attacks (filtration, noise, etc.). For the well known JPEG/JPEG2000 attacks, the watermark survived at a quality factor Q≤1 with JPEG compression, and at compression ratios up to 500:1 with JPEG2000.

Introduction

Digital watermarking is a well-known technique for copyright protection. In general, a digital watermark is a code or owner information that is embedded inside an image. It acts as a digital signature, giving the image a sense of ownership or authenticity [1], [2]. Robust watermarking embeds information data within the image with an insensible form for human visual system, but in a way that protects from attacks such as common image processing operations [2], [3]. The goal is to produce an image that looks exactly the same to a human viewer, but still allows its positive identification in comparison with the owner's key if necessary.

Digital watermarking technique must satisfy the following four properties: firstly, the embedded watermark must not spoil the quality of the image and this is mostly measured by the PSNR of the watermarked image. Secondly, the watermark should be perceptually invisible. Thirdly, the watermark should be robust against modification that can be done by removing the original watermark and replacing another one, or by inserting another watermark in such a way that both are detectable. Finally, the watermark should be robust to common signal processing operations and geometric distortions in order to aid different applications [4]. For high quality and low visual distortion, the watermark must modify the information contained in the original image with as minimum as possible modification to the pixel value. This change is the watermarking error introduced by the watermark insertion and is usually measured by the mean square error (MSE) [1]. In case of lossless watermarking, the watermarking error is set to zero. It is often desirable to retrieve the watermark without referring to the original image; that is known as blind watermarking [5], [6].

In order to increase the robustness of the watermarking scheme many approaches were discussed in literature and were mainly based on the discrete cosine transform (DCT) or the discrete wavelet transform (DWT). Cox and Miller [1] explained the principles that govern the watermarking procedure and watermarking applications. Thus, various properties of watermarks, such as how they respond to common signal transformations or deliberate attacks were discussed. The authors gave an example of a basic class of watermarking methods in a mathematical model, and used restricted and unrestricted keys to achieve the secrecy of the watermark. Wang et al. [2] discussed a blind wavelet-based algorithm for ownership verification of digital images. Here, the host image is transformed using multilevel DWT construction, the sub-band selected to embed the watermark is chosen from the middle frequency sub-bands, which enables minimum perceptual error and robustness against frequency based filtering. The binary watermark is transformed into real numbered value by using a rotational matrix; then, real numbered watermark is weighted by suitable coefficient that gives enough power to the watermark. The insertion process is done by changing the selected sub-band by the weighted real valued watermark. However, the capacity of this method will not be large enough to contain the owner data, moreover; it is easy to extract the watermark once the host sub-band is known.

Houng et al. [7] investigated a wavelet-based watermark casting scheme for blind and non-blind watermark retrieval. They developed an adaptive watermark casting method to first determine significant coefficient wavelet sub-bands and then select a couple of significant wavelet coefficients in theses sub-bands to embed watermarks. This method gives a perceptual weighting for different significant wavelet coefficients, and sets a limit on the bound of the fidelity loss after watermark casting [8]. The blind detection is more complicated because it depends on the expected error that the watermark has introduced to each significant wavelet coefficient. This method resists well against compression and filtering with large capacity. However, for blind method it is neither practical nor reliable with respect to the non-blind method because it requires a large capacity to store the original image.

Wang and Lin [3] proposed a wavelet-tree-based blind watermarking scheme for copyright protection. They benefited from [9], where the watermark is embedded in the wavelet coefficients in such a manner that each watermark bit is embedded by quantizing a single wavelet coefficient out of a set of coefficients corresponding to a particular spatial region. In [3], the wavelet coefficients of the host image are grouped into what is called super trees and the watermark bit is embedded by quantizing all the coefficients in super trees. The trees are quantized until enough difference is achieved between these trees. The difference between quantized and un-quantized trees carries the watermark information required to extract the watermark. The proposed method resist to the most common attacks like filtering and JPEG compression. The main shortcoming of this method is attributed to the quantization process that requires a high computational cost. Moreover, the proposed encoder does not consider the mutual relations between these trees to enhance the PSNR value of the watermarked image.

Several alternatives have been suggested to apply watermarking techniques to color images. One commonly used approach consists of marking the image luminance component only, which makes the extension of grayscale techniques straightforward. The most relevant drawback to luminance watermarking is in the fact that luminance information is more perceptible to modifications than the chrominance. Other approaches, based on using the color components, were also proposed. For example, Kutter et al. [10] used the blue channel to embed the mark, since it is claimed that the Human Vision System (HVS) is less sensitive to the blue band in the RGB space. This approach is similar to the grayscale schemes, since the watermark is embedded only in one layer of the RGB image. Piva et al. [11] have made an extension to the approach considered in [10]. Here, the three RGB channels were used to embed the watermark. The selection of the magnitude of the embedded watermark has taken into account the HVS sensibility to the spectrum: the magnitude of the watermark is 10 times more significant for the blue component compared to the green, and 5 times for the blue component compared to the red one.

Fleet and Heeger [12] proposed a quite different approach which embeds a sum of sinusoids in the yellow–blue channels of the opponent-color representation. Here, the L*a*b space is extended to the spatial CIELAB (S-CIELAB), which includes the spatial structure of the image. A more recent work can be found in [13], where a quaternion space is considered, then, the quaternion Fourier transform is applied to embed the watermark in the transformed domain.

Tsai et al. [14] proposed a new color image watermarking scheme based on color quantization. The proposed watermark embedding procedure is imposed on the pixel mapping process of the color quantization process. When the pixel mapping procedure is performed, the watermark is embedded at the same time, i.e., the watermark is embedded into the color index table. The proposed watermark extracting procedure is also imposed on the image decoding procedure in color quantization. However, the palette plays an important role on the quality of the quantized image and affects the robustness of the proposed scheme.

Dietl et al. [15] implemented wavelet filter parameterization as a means to add security to wavelet-based watermarking schemes. It turned out that a combination of filter parameterization and non-stationary wavelet decomposition achieves a keyspace of cryptographically reasonable size. Also, robustness against JPEG and JPEG2000 compression was found to be on an equal level as compared to the use of standard wavelet filters.

Al-Otum and Tabaa [16] proposed a family of modified wavelet-based watermarking techniques that is based on the improved pixel-wise (PW) watermarking scheme [17]. The basic proposed algorithm is based on selecting specific locations in the three detailed sub-bands of the first level of the DWT decomposition of the image. However, the proposed techniques can only detect the existence of the watermark, and have no extraction capability since they only detect the existence of the watermark.

Finally, Al-Otum and Samara [18] proposed an adaptive blind wavelet-based watermarking scheme using tree mutual differences that is based on exploiting mutual differences between grouped coefficients of the wavelet-trees in the grayscale domain. The encoder adaptively searches for the bit host difference in such a manner to minimize the embedding error. However, the technique is proposed to be applied to grayscale images, and does not exploit the color inter-channel relations in the bit-host construction procedure.

In this paper, a robust wavelet-based blind technique for color image watermarking is proposed. The trees are generated individually for each color component; each watermark bit is embedded using two trees from two different color components. The bit insertion process modifies the difference (energy) between the selected two trees in such a manner that this difference carries the bit sign with sufficient energy which insures the robustness. As a result, there will be enough space to select the suitable bit host that minimizes the watermarking error which, then, enhances the visual quality of the watermarked image. This method simplifies the detection process, by extracting the sign of the difference that was chosen to represent the watermark bit.

The arrangement of this paper is as follows: in Section 2, we formalize the main objectives of the paper, introduce the proposed robust wavelet-based blind technique for color image watermarking. Then, in Section 3, simulations are carried out to check for the imperceptibility, capacity and robustness of the proposed scheme. Finally, conclusions are given in Section 4.

Section snippets

General idea

In color (RGB) images, each color pixel is constructed using 24 bits with 8 bits per color component (R, G, and B). Each color is represented by a unique combination of the three quantities; any change of these quantities will change the color itself, even though, this combination is preserved when the image is transformed into the wavelet domain, the relation between the wavelet coefficients, corresponding each pixel, is kept unique. This uniqueness will be used to construct a robust

Simulation and results

In this section, the properties of the proposed CW-BHDS algorithm are tested using a wide variety of color images with the size of 512×512 pixels and 24 bits/pixel resolution. Examples of the implemented test images are: Lena, Baboon and Airplane. The implemented DWT is a 4L-DWT with Daubechies-1 filtering kernel.

Typically, three main measures are commonly used to assess the performance of the watermarking scheme: (1) Imperceptibility, (2) Capacity, and (3) Robustness. Next, the performance of

Conclusions

In this work, a blind watermarking technique based on the so called wavelet-trees is introduced, and denoted as Image Watermarking based on wavelet-tree bit host difference selection (CW-BHDS). Unlike many existing algorithms, that consider colored image as independent R,G, and B layers, the CW-BHDS treats the color pixels as one unit and exploits the significant features and relations between pixel color components. An important feature of the CW-BHDS algorithm is that it is based on

References (21)

  • P. Tsai et al.

    A color image watermarking scheme based on color quantization

    Signal Processing

    (January 2004)
  • W. Dietl et al.

    Protection of wavelet-based watermarking systems using filter parameterization

    Signal Processing

    (October 2003)
  • H. Al-Otum et al.

    Adaptive color image watermarking based on a modified improved pixel-wise masking technique

    Computers and Electrical Engineering Journal

    (September 2009)
  • I. Cox et al.

    Digital Watermarking, Morgan Kaufmann Series in Multimedia Information and Systems

    (2001)
  • Y. Wang et al.

    A wavelet-based watermarking algorithm for ownership verification of digital images

    IEEE Transactions on Image Processing

    (February 2002)
  • S.-H. Wang et al.

    Wavelet tree quantization for copyright protection watermarking

    IEEE Transactions on Image Processing

    (February 2004)
  • D. Kundur et al.

    Diversity and attack characterization for improved robust watermarking

    IEEE Transactions on Signal Processing

    (October 2001)
  • J. Eggers, J. Su, B. Girod, Robustness of a blind image watermarking scheme, available at:...
  • I. Cox, M. Matt, J.A. Bloom, Watermarking applications and their properties, available at:...
  • H.-J.M Wang et al.

    Wavelet-based digital image watermarking

    OPTICS EXPRESS 4917

    (December 1998)
There are more references available in the full text version of this article.

Cited by (57)

  • Robust and blind watermarking technique in DCT domain using inter-block coefficient differencing

    2016, Digital Signal Processing: A Review Journal
    Citation Excerpt :

    Robustness of watermark is acceptable only if the extracted watermark has very small number of distorted bits [4]. Further watermark must withstand various attacks as well as watermark insertion should not spoil the image quality to a large extent [5]. A watermark can be embedded into a host image either by modifying the pixel intensities in spatial domain or by modifying the coefficients of the transform domain of the host image [6–10].

  • Robust watermarking against geometric attacks using partial calculation of radial moments and interval phase modulation

    2014, Information Sciences
    Citation Excerpt :

    Robust watermarking can be used to trace copies or to implement copyright protection methods. Some watermarking algorithms are published in order to satisfy this robustness requirement [2,7,13–15,18]. For robust watermarking schemes, the embedded watermark must be robust against a variety of possible attacks.

  • A novel blind color images watermarking based on SVD

    2014, Optik
    Citation Excerpt :

    Digital watermarking is to embed the important information into the digital data (audio, image, video, and text). According to the processing domain of the host image, the existing techniques on image watermarking may be roughly divided into two categories, i.e., frequency-domain and spatial-domain methods [1]. Although more information for embedding and better robustness against the common attacks can be achieved through frequency-domain method, the computational cost is higher than the ones based on spatial domain.

  • Color image blind watermarking scheme based on QR decomposition

    2014, Signal Processing
    Citation Excerpt :

    With the rapid development of Internet and multimedia technology, how to protect the copyright of products from the illegal usage has been receiving more and more attention. Among the published techniques of copyright protection, digital watermarking is considered as a powerful method [1,2]. The feature of digital watermarking is to allow for imperceptibly embedding watermark information in the original multimedia data.

View all citing articles on Scopus
View full text