Digital image visible watermarking is a type of digital watermarking technology
(Phadikar et al., 2007; Shirali-Shahreza
and Shirali-Shahreza, 2008; Islam et al., 2008;
Prasannakumari, 2009; Abdulfetah
et al., 2009) used for protection of publicly available images.
It offers a wide range of potential applications in many fields such as digital
libraries, digital TV, e-commerce and video on demand via the internet. Different
from invisible watermarking algorithms (Liang et al.,
2011; Olanrewaju et al., 2011; Shilbayeh
and Alshamary, 2010; Zeki et al., 2011; Yang
et al., 2010), the visible watermarking schemes embed watermarks
into digital contents translucently by overlaying recognizable but unobtrusive
copyright patterns (watermarks). The watermark should be noticeable yet must
not significantly obscure the image details beneath it. A well-designed visible
watermarking technique should retain all details of host contents. At the same
time, the watermark must be difficult or even impossible to remove. Many visible
watermarking approaches have been proposed in literature. Braudaway
et al. (1996) presented an earlier visible watermarking approach
in spatial domain to mark the digitized pages of manuscript from the Vatican
archive. Mohanty and Ramakrishna (2000) designed a visible
image watermarking in DCT domain by employing the texture sensitivity of the
Human Visual System (HVS) and modifying the DCT coefficients of the host image.
Hu and Kwong (2001) gave a scaling factors calculation
strategy based on the effect of luminance and local spatial characteristics
and implement a wavelet based visible watermarking technique. Biao-Bing
and Tang (2006) proposed a visible watermarking algorithm in discrete wavelet
transform domain by using a contrast-sensitive function. This approach provided
good watermark visual effect on watermarked images. Min-Jen
et al. (2011) presented an Adaptive visible image watermarking approach
in wavelet transform domain by incorporating contrast sensitivity. The strength
of the watermark within the wavelet decomposition sub-bands is adjusted according
to this sensitivity. The proposed method has excellent robustness and transparency.
Zeng and Wu (2010) described a spatial image adaptive
visible watermarking scheme by considering both the host image and visible watermark
features. However, all the above-mentioned schemes the unique characteristics
of document images and they are not applicable for document images.
To protect copyright of digital document images in digital library, we propose an adaptive visible watermarking scheme for document images by exploiting the luminance and edge sensitivity of document images.
CALCULATION OF SCALING FACTOR AND EMBEDDING FACTOR
To obtain a well trade-off between the watermark embedding energy and the quality of perceptual watermark translucence during visible watermark embedding, it is important to exploit the visual perception host images.
Human visual perception is more sensitive to distortion in blocks with mid-intensity values than that of low intensity blocks as well as high intensity blocks. In fact, sensitivity drops parabolically as the gray value drifts on both sides. So, luminance masking used can be described by the formula:
where, θ(x, y) denotes luminance masking, I (x, y) is the gray value of
a pixel with coordinates x and y,
is the mean gray value of the host image, μ'max is the maximum
value of μ' (x, y).
The edge is an important image structure. Human eyes are very sensitive to the noise in image edge areas. Since, Image edge areas have greater variance value while smooth image areas have smaller variance value, the edge feature can be reflected by variance δ (x, y) (note that: the variance is calculated in an image neighbor with size 3x3). The edge visual factor (λ(x, y)) is defined by the normalized variance as:
where, δmax is the maximum value of δ (x,y).
Document images have a distinctive feature that they are composed of text areas and non-text areas (background). Text area is black or near black, but large background areas, usually white or near white. These areas can tolerate large gray value change. This means that the scaling factor a (x, y) will monotonically decrease with luminance masking θ (x, y) while the embedding factor β (x, y) monotonically increase. Moreover, the edge areas locate on the junctions of text and non-text areas in document images. Border region should avoid significant distortion during watermark embedding. For simplicity, we can assume the scaling factor a (x, y) to be directly proportional to the edge visual factor λ (x, y). The variation of the embedding factor β (x, y) with respect to the edge visual factor λ (x, y) is the reverse of that of the scaling factor a (x, y).
According to the above considering, a (x, y) and β (x, y) can be computed as follows:
Finally, to achieve a flexible and controllable visible watermark embedding, we normalize the scaling factor and the embedding factor to a closed interval [a, b] (where, 0<a<b<1) and obtain normalized a (x, y) and β (x, y) by the following formula:
where, function max() and min() return the maximum and minimum value of an expression, respectively.
VISIBLE WATERMARK INSERTION FOR DOCUMENT IMAGE
In visible watermark embedding for document images, our scheme considered the
unique characteristics of document images by exploiting the luminance and edge
masking characteristics. The whole visible watermark embedding process is described
||Watermark image W, original document image I
||Watermarked document image I'
||Read the original document image I and the watermark image
||Compute the scaling factor a(x,y) and the embedding factor a(x,y) for
each pixel according to Eq. 4
||Choose an image pixel of the original document image for visible watermark
||The corresponding watermark message bit W(x,y) is embedded by modifying
pixel value I(x,y) of document image I as follows:
||Repeat step 3-step 4 until all the watermark bits are embedded into the
host document images and finally the watermarked image I' is generated
In This study, some gray-scale document image with size 512x512 are adopted
as test images and watermark images are grayscale images with recognizable pattern
which have the same size as the host document images. Parameters a and b in
Eq. 4 value 0.1, 0.9, respectively.
Figure 1 and 2 show the watermark image and four test document images.
|| Watermark image
||Test document images
|| Watermarked images generated by our scheme
||Watermarked image visual quality (PSNR) comparison between
the proposed method and Zeng and Wu (2010) (dB)
Figure 3 shows the visible watermarking experimental results
carried out on test document images shown in Fig. 2 and watermark
image as seen in Fig. 1. In addition, we have also compared
the proposed scheme with Zeng and Wu (2010) and the
detailed comparison experimental results are listed in Table 1.
From Fig. 3, we know that the watermarked images have good
visual quality and that the quality of perceptual watermark translucence is
pleasant. PSNR comparison results in Table 1 illustrates that
our approach obtains about 2 dB higher PSNR value of watermarked images than
Zeng and Wu (2010) on average.
||Robustness against removal attacks, the watermarked image
(a-f) are obtained by histogram equalization, mean filtering with 3x3 window,
image binarization with threshold 240, averaging 25 different watermarked
image versions, JPEG compression with quality factor 10 and JPEG 2000 compression
with compression ration 60:1, respectively
The proposed scheme can get an optimal trade-off between the watermark embedding
strength and the perceptual watermark translucence. This is because that our
scheme considered the document image visual perception especially by combining
luminance masking and edge masking characteristics.
In addition to watermarked imaged quality and perceptual watermark translucence, the visible watermark imposed on host document images must be difficult to remove. That is to say, visible watermarking schemes should be robust against common attack operations such as image compression, image enhancement, image filtering and collusion attack and so on. Take Fig. 2a as an example, the illustration of robust ness experimental results is shown in Fig. 4. We can see that the watermarked document images have satisfactory visible watermark translucence and that the proposed document image visible watermarking technique is robust against ordinary signal processing attacks.
An adaptive visible watermarking technique for document images has been proposed in spatial domain by exploiting the unique visual perception of document images. The scaling factor and the embedding factor are adaptive to document image content due to the combination of luminance masking and edge masking. The proposed scheme achieves best trade-off between the watermark strength and the perceptual watermark translucence and it obtains pleasant document image quality, watermark visibility and robustness to common removal attacks. It will have widely applications in digital media fields such as digital library, digital cinema and digital publishing.
This study was supported by Youth Foundation of Research Program of Humanities and Social Sciences of Chinese Ministry of Education (10YJC870014), National Natural Science Foundation of China (61073191), Hunan Provincial Philosophy and Social Sciences Foundation of China (11YBA090), Hunan Provincial Natural Science Foundation of China (10JJ6090), Scientific Research Fund of Department of Science and Technology of Hunan Provincial, China (2011GK3140), Science and Technology Innovative Research Team in Higher Educational Institutions of Hunan Province (2010212).