Contourlet-based image adaptive watermarking
Introduction
Copyright protection of digital image is one of challenging issues imposed in ubiquitous media. Digital watermarking for images has been identified as a possible solution to this challenge, and has become an area of increased research activity over the last decade. Image watermarking is the process of embedding the owner's mark into image data so that intellectual property rights can be identified. In general, a watermarking scheme shall satisfy two properties. First, the embedded watermark does not visually distort the image. The second property is that the watermark is difficult for an attacker to remove. It should be robust to common signal processing and geometric distortions, such as low-pass filtering, compression, noise, affine, and rotation.
Watermarking may be performed in either image domain or transform domain, such as discrete Fourier transform (DFT), discrete wavelet transform (DWT), and discrete cosine transform (DCT). Schyndel et al. [1] proposed to insert a watermark by changing the least significant bit of some pixels in an image. Bender et al. [2] described a watermarking approach by modifying a statistical property of an image. Recent contributions have shown that image adaptive watermarking schemes can be successfully implemented using image transforms with spatial-frequent local properties. Kang et al. [3] embedded watermark in the coefficients of the LL subband in the DWT domain while a template is embedded in the middle frequency components in the DFT domain to achieve the robustness to both affine transform and JPEG compression. The approach implemented as a DCT–DWT dual domain algorithm and applied for the protection and compression of cultural heritage imagery is proposed by Zhao et al. [4]. In order to guarantee the quality of a watermarked image not to degrade visually, watermark should be embedded in high-frequency (HF) components. However, it makes the scheme vulnerable to attacks such as compression and low-pass filtering. Performance gain can be obtained by exploiting the characteristics of the human visual system (HVS) in the watermarking scheme [5], [6]. They compensated for the lack of robustness in the HF components by increasing the watermark strength to its maximum, while preserving the imperceptibility thanks to the visual masks.
Conventional watermarking schemes embed watermark into a certain scale subbands in transform domain (either HF subbands or low-frequency (LF) subbands), thus they can only resist the attack of a particular kind of image processing. Considering a watermarking scheme that embeds watermark into HF subbands, the watermark will be removed easily when the watermarked image is attacked by image processing methods which destroy the HF of the image, although it is robust to the LF filtering. Therefore, most of conventional watermarking schemes are semi-robust in general.
In this paper, we propose a contourlet-based image adaptive watermarking (CIAW) scheme. In the CIAW scheme, although the watermark is embedded into the largest detail subbands (the highest-frequency subbands), it is likely to be spread out into all subbands when we reconstruct the watermarked image, due to the special transform structure of Laplacian pyramid (LP) [7]. Because the LF subbands of the watermarked image contain the watermarking components, the proposed CIAW scheme is very robust against various HF attacks, such as low-pass filtering, quantization, and compression, which will destroy the HFs of the image. On the other hand, some watermarking components can be preserved at the HF subbands. Thus, the CIAW scheme is expected to be also robust to the LF attacks, such as gamma correction, histogram equalization, and cropping, which will destroy the LFs of the image. Consequently, the proposed CIAW watermarking scheme is robust to the widely spectral attacks resulting from both the LF image processing and the HF image processing.
The watermarking detection algorithm corresponding is proposed to decide whether the watermark is present or not by exploiting the unique transform structure of LP. It checks the correlation between the watermarked image and the spread watermark in all subbands of contourlet domain. Our detection algorithm is superior to the conventional detection algorithms because it can exploit the correlation of the watermark and all watermarking components in different subbands of the watermarked image fully.
The rest of this paper is organized as follows. In next section, the contourlet transform (CT) is described in detail and the characteristic of LP is analyzed from the viewpoint of watermarking embedding. Section 3 presents our CIAW scheme together with its corresponding watermarking detection algorithm. In Section 4, the simulation results are presented. Lastly, Section 5 concludes this paper.
Section snippets
Contourlet transform
Contourlet was proposed by Do and Vetterli in 2001 [8], [9]. It can efficiently represent contours and textures of an image. Contourlet is a double filter bank (FB) structure for obtaining sparse expansions for typical images with smooth contours, where the LP is used to capture the point discontinuities firstly, then followed by a directional filter bank (DFB) to link the point discontinuities into linear structures. The overall result is an image expansion using basic elements like contour
Contourlet-based image adaptive watermarking (CIAW) scheme
The watermark W to be embedded can be arranged as a set of matrices Ws,d(i, j) with the size MW×NW and the pseudo-random binary values {−1, 1}. The indices s and d indicate the scale and the direction of contourlet subband being embedded by watermark, respectively. Embedding watermark into the contourlet subbands of an image is accomplished according to where Cs,d(i, j) and C′s,d(i, j) are the original contourlet coefficient and the watermarked
Experimental results
To validate the invisibility and the robustness of the CIAW scheme, we conducted the experiments on the different images (“Lena”, “Barbara”, and “Peppers” images of size 512×512) and simulated some image processing operations, which may remove the inserted watermark. The 5/3 bi-orthogonal filters are used for both the multiresolution pyramidal filtering and the directional decomposition. Firstly, we transform the image into contourlet representation with S=3. Subsequently, we embed the
Conclusion
We have proposed the new watermarking scheme, namely CIAW, which enjoys both the invisibility and the robustness. Correspondingly, a new watermark detection algorithm has been proposed for the CIAW by fully exploiting the characteristic of CT, which can detect the watermark efficiently from the watermarked image under severe attacks. Experimental results have shown that the proposed scheme is very robust to common attacks such as low-pass filtering, compression, noise, affine, and rotation.
References (14)
- et al.
A digital watermarking
IEEE International Conference on Image Processing (ICIP’1994)
(1994) - et al.
Techniques for data hiding
IBM System Journal
(1996) - et al.
A DWT-DFT composite watermarking scheme robust to both affine transform and JPEG compression
IEEE Trans. Circuits Syst. Video Technol.
(August 2003) - et al.
Dual domain watermarking for authentication and compression of cultural heritage images
IEEE Trans. Image Process.
(March 2004) - et al.
Image-adaptive watermarking using visual models
IEEE J. Sel. Areas Commun.
(May 1998) - et al.
Improved wavelet-based watermarking through pixel-wise masking
IEEE Trans. Image Process.
(May 2001) - et al.
Framing pyramids
IEEE Trans. Signal Process.
(September 2003)