Figure 1. Embedding and Retrieval of the watermark

ABSTRACT

In this paper a novel algorithm based on wavelet transform is proposed for digital image watermarking. The algorithm is primarily combinations of TWT (discrete wavelet transform) and modified throsholding scheme to search exact coefficient for watermark embedding. The proposed algorithm gives good results of watermark retrieval also tolerate the attack of low pass filter and high pass filter, it is more robust against scaling. Retrieval of watermark is in the absence of original image and improves the fidelity.

KEY WORDS

Invisible Digital Watermarking, Discrete Wavelet Transform, Sub band Threshold Scheme.

1. Introduction

Digital watermarking is considered as the imperceptible, robust, secure communication of information by embedding it in and retrieving it from the digital images. Robustness and fidelity requirements are the key issues in watermarking. The fidelity and robustness constraints often conflict. Here, Images watermarked with non-robust methods exhibit an unacceptable loss of fidelity. When the watermarked image is not distorted between the times of embedding and detection it is a robust watermark.

Ideal properties of a digital watermark are: Watermark must be invisible, no degradation of image after embedding watermark, more fidelity, more robust, statistically invisible. Watermark extraction should be fairly simple and accurate, must be strongly resistant to detection, tolerate different attack and no loss of image quality. Watermark is the proof of ownership and authentication

In [1], DCT and patchwork method is implemented, it observed that, the patchwork is in particular portion, in case this portion is cropped then watermark retrieval is very poor but by our proposed method watermark is distributed throughout the image. The method discussed in paper [2] maintained that watermarked image was distorted in turn by filtering but in the proposed method correlation coefficient remains same when PSNR is changed for the attack of high pass filter and low pass filter.

In earlier method [3] if the scaling factor changes the fidelity of the watermarked image changes

which leads to degradation of image. Also, some types of image processing attacks are not tested.

Existing method needs the improvement in retrieval of watermark and the fidelity is not good. The novel contributions of the proposed method are, the watermarked image fidelity is independent of scaling factor and it gives good improvement in retrieval of watermark, and also tested against the attacks of high pass filter and low pass filter.

The proposed method uses the wavelet transform because it reduces the visible degradation and has been shown to give good compact representation of the image texture. This suggests that it may have powerful watermarking properties. Proposed method does not require the host image to extract the watermark. The advantage, of the DWT approach is that embedding the watermark sequence at the various resolution levels. This approach provides a simultaneous spatial localization and frequency spread of the watermark within the host image. In addition, the watermark merging process is adapted as it depends on the local image characteristics at each resolution level, and is robust as it embeds the watermark more strongly into more salient components of the image, when the watermark embedded in silent components there was no degradation of the image and we get good fidelity. The combined result of these factors makes the proposed method attractive.

To overcome the problems of method proposed in [3], the proposed algorithm recommends for maximum value threshold subband instead of taking half of the maximum threshold value. For the remaining sub bands instead of one forth of the threshold value, half the threshold is used for embedding watermark. The use of proposed values leads to increase in fidelity, robustness and accuracy of the retrieval of watermark. For visual comparative study images are shown in the figure 3.

In the section 2 we discussed some related work needed to prove good digital image watermarking. Section 3 introduces the proposed method of Watermark embedding and retrieval for digital watermarking. In section 4 we compare the results of existing algorithm with the proposed algorithm. We provide concluding remarks and discussion in section 5.

2. Related Work

It is very economical and faster to send still images in digital form via the World Wide Web rather than the hard copies sent by post. But while sending on the World Wide Web there is possibility of the misuse of information such as illegal copying, duplication, etc. To avoid this some watermarking methods are invented. This watermark is designed to identify both the source of a document as well as its intended recipient. It is also used for the purpose of proving legal ownership and to detect copyright violations Watermarking ensures the security of patents. A watermark is an invisible mark (binary code) embedded in an image that is detected when the watermarked image is compared with the original or it may be retrieved directly from the image available for investigation. To design the algorithm for watermarking the three most critical requirements are fidelity, robustness to attack, and watermark capacity [4].

The owner of images like publishers, artists, and photographers, however, may be unwilling to distribute pictures over the Internet due to a lack of security; images can be easily duplicated and distributed without the owner’s consent. In case of suspected copies of pirated versions it is needed to check the ownership [5].The presence of watermark does not visibly degrade the document, but can be easily detected by the owner. Popularity of different techniques for processing of watermarks in text, images, video and audio have increased in recent years [6].

Most of the recent work in watermarking can be grouped into two categories: spatial domain methods [7] and frequency domain methods [8]. In these methods discrete wavelet domain is used. A multi resolution data fusion approach, in which the image and watermark are both transformed into the discrete wavelet domain, is employed. There is a current trend towards approaches that make use of information about the Human Visual System (HVS) to produce a more robust and invisible watermark but it needs the original image to extract the watermark [9]. In [10] different transforms are used to embed watermark like DCT, Walsh transforms, wavelet transform and FFT, it shows robust watermarking but these methods need the original image to retrieve the watermark. Also, it is not clearly mentioned whether the watermark was sustained for different attacks.

In [11] uses quadtree to protect the information in watermarked image, to make simple embedding and retrieval of watermark and to provide robustness for some type of attacks. Robust spread spectrum data hiding and a linear model is used to increase the data hiding capacity [12].In [13] multiplicative watermarking method is used, filter and circular harmonic functions compute the required edge image and ridgelet transform is applied to get most significant coefficients to get robust and transparent embedding of watermark and also some types of attacks are tested. This method breaks down when the ‘Stirmark’ random geometric distortions are considered. In [14] multiple watermarks are embedded into different compression domains to protect the ownership of the original images. It gives robustness for surviving different attacks, multiple watermarks are embedded into the vector quantization domain as well as for hiding the secret keys associated with the watermark in the transform domain, and extraction of the watermark does not need the original image. In [15] original image is not required during the ownership verification. Statistical modeling techniques are applied to the analysis of two watermarking schemes, one of them defined in the spatial domain, and the other in the direct cosine transform (DCT) domain to get considerable improvements in the performance. Low power, real time, reliable and secure watermarking systems, is developed through hardware implementations, for this FPGA based implementation of an invisible spatial domain watermarking encoder is applied [16]. For oblivious watermarking [17] multiple description frameworks i.e. a spread-spectrum watermarking algorithm for DCT based multiple descriptions is described. Lattice coding approach is used for effective analysis and design for semi fragile watermark to achieve secure robust and fragile watermarking [18].

Analytical approach of pilot based synchronization is used in this Levenberg-Marquardt’s for nonlinear least-square estimation and shows how an estimate of the geometrical transformation parameters can be obtained it increases the capacity of hidden information [19].

A robust watermarking based on fuse mark using multiresolution data fusion to hide high energy logo in silent image components to decrease the problem of false negative detection without increasing the false positive detection rate [20].

3. Proposed Watermarking Technique

The process of digital watermarking involves the modification of the original data image to embed a watermark containing key information as a threshold of sub band. DWT is applied to the original image to separate it into different bands. At each level there are four subbands of approximately equal bandwidth on a logarithmic scale. The watermark is then embedded as the binary sequence of 1 and 0.

3.1. Exact Coefficients of Watermark Embedding

The proposed algorithm employs significant coefficient search (SCS) method [1]. A binary watermark sequence consisting of 1 and 0 embedded into the wavelet coefficients depends on the threshold value of the each sub band. In this method different sub bands are given proper weighting for different significant wavelet coefficients. The sub band consists of number of bit plane layers and in which bit plane layer watermark is to be embedded depends upon the threshold value of each sub band. T, threshold value of each sub band‘s’, is calculated as. For the subband of highest threshold value use Ts=T and for other sub bands use the value of Ts=T/2. Where C_{max ,s} is effect is good i. e. fidelity effect is good and it is independent of the

change in scaling factor alpha, at the same time accuracy of watermark retrieval is maximum.

3.2Watermark Embedding and Retrieval

In the proposed method the cameraman gray level image of dimension 256 X 256 is used.

3.2.1. Watermark Embedding:

Invisible watermark embedding is shown in figure1. Basic steps of embedding the watermark into the original image X (i, j) are:

Step 1: Decompose the original image X ( i, j) into three resolution levels. In each level there are three detailed and one approximate image.

Step 2: Check the threshold value of the subband and embed the watermark in it by processing

Where,

is the coefficient of the watermarked image,

X is the original coefficient, and are scaling factors,

varies from 0 to 1 and is calculated from threshold value of each subband .

T_S is the current threshold of subband s in the j^th bit plane,

W_K is the k^th watermark element in a watermark sequence of length N_w, and W_K is the watermark

sequence 0 and 1.

In the previous method as increases the fidelity of watermarked image decreases but in proposed method has no effect on the fidelity of the watermarked image. Hence, there is no degradation of watermarked image.

Step 3: Apply the inverse discrete wavelet transform (IDWT) to compute the watermarked image (i, j).

3.2.2. Watermark Retrieval

There is no need of original image to retrieve the watermark. To extract the watermark from watermarked image first apply the DWT to the watermarked image up to third level. A sequence of the watermark is extracted from the wavelet coefficients of each sub band of the watermarked image, the threshold scheme same as described in the section 3.1, the difference is that instead of adding a watermark sequence we have to subtract the sequence according to the values of the wavelet coefficients. Extracted watermark were compared with the original watermark to compute the retrieval accuracy and extracted watermark were correlated with the original watermark to compute the correlated coefficients . The proposed technique proved to be more robust than the DCT method [6].

4. Results

Figure 2. shows a standard image without a watermark.

Figure 3 show same image watermarked using earlier method i.e. significant coefficient search [1] and the modified proposed method. It can be seen that the image is degraded, for scaling factor alpha equal to one, for the earlier method. The watermarked image displayed by proposed method visibly looks like the original image.

The results of accuracy of the retrieval of watermark and the PSNR (probability of signal to noise ratio) for alpha=0.1 and alpha=1 without any type of attack are tabulated in Table 1. It is observed that the accuracy of the retrieval of the image is maximum for the proposed method.

Table 1: Comparative Values of Different Parameters.

Parameters
	Exist method	Proposed method	Exist ethod	Proposed method
Retrieval efficiency	99.86%	100%	99.67%	100%
PSNR	42.58	42.44	22.58	22.44

Table 2: Comparative Values of Different Parameter after the attack on watermarked image.

Parameters	Low Pass Filter
	Exist method	Proposed method		Exist method		Proposed method
	Retrieval efficiency	96.99%	99.22%		96.43%		99.46%
PSNR	42.58	42.44		22.58		22.44
CRC	0.9065	-0.3055		0.9791		0.1101
Parameters	High Pass Filter
	Exist method	Proposed method	Exist method		Proposed method
	Retrieval efficiency	81.39%	98.49%	81.15%		98.44%
PSNR	42.58	42.44	22.58		22.44
CRC	-0.9703	-0.7726	-0.9288		-0.7664

.

From the figure 3.b it is also evident that the fidelity is very good. The accuracy and the PSNR and CRC (correlation coefficient) results of retrieval of the watermark, from attacked image, for alpha=0.1 and alpha=1 after attack of high pass filter and low pass filter are provided in table no.2. It is seen that the proposed method gives good results.

From plots shown in Figures 4.a it can be seen that robustness and fidelity of the watermarked image is more by the proposed method after the attack on the watermarked image by low pass filter and high pass filter. Also CRC is near about same for the attack of high pass filter or low pass filter attack shown in figure 4.b. Also it is observed that the watermark retrieval accuracy is higher for the attack of high pass filter or low pass filter as compare to the previous method it is shown in figure 4.c.

4. Conclusion and future scope

In this paper we propose a robust method for still-image watermarking based on concepts of discrete wavelet transform, subband thresholding of the image and embedding watermark in the form of binary sequence. Advantage of the proposed technique over earlier approach is that firstly it has good fidelity, secondly it is highly robust, thirdly it provides high retrieval accuracy of watermark, and finally it also gives good results for low pass filtering and high pass filtering attack.

Some of the advances discussed above are in their infancy, and much interesting work remains to be done. In some cases, we believe that significant results may be imminent, which makes this area exciting. In other cases, we do not see any breakthroughs on the horizon, but significant results would increase the suitability of watermarks for a wider variety of applications, and are therefore worth further study. It would be interesting to go for further attacks like compression, rotation etc. and to study effects on watermarking of images.

6. Reference:

[1] Hiroyuki Kii, Junji Onishi, Shinji Ozawa, “The Digital Watermarking Method by Using both Patchwork and DCT,” Multimedia Computing and Systems, IEEE, Vol.1,pp895-899,1999.

[2] J.Ohnishi and K.Matsui, “Embedding a seal into a picture under orthogonal wavelet transform,” in Proc. Int. Conference in image Processing, Vol.2, 1994.,86-90,

[3] Houng_jyh, Mike Wang,Po_Chyi Su and C._C.Jay Kuo, “Wavelet_ Based digital image watermarking,” OPTICS EXPRESS, Vol.3,No.12, December 1998, 491-496.

[4] R.B. Wolfgong, C.I.Podilchuk, and E.J. Delp, ‘Perceptual watermarks for digital images and video,” in Proceedings of the IEEE,Vol.87,1999,1108-1126.

[5] G. Voyatzis and I. Pitas, “The use of watermarks in the protection of multimedia products,” in Proceedings of the IEEE, Vol.87, 1999,1197-1207.

[6] I.F. Hartung and M.Kutter, “Multimedia watermarking techniques,’ in Proceedings of the IEEE,Vol.87,1999,1079-1107.

[7] N. Nikolaidis, I. Pitas, “Robust Image Watermarking in the Spatial Domain,” Signal Processing, Vol. 66, No. 3,1998, 385-403,.

[8] D. Kundur, D. Hatzinakos, “Digital Watermarking using Multiresolution Wavelet Decomposition”, Proc. IEEE Int. Conf. On Acoustics,

Speech and Signal Processing, Seattle, Washington, Vol. 5, May 1998,2969- 2972.

[9] Deepa Kundur and Dimitrios Hatzinakos, “A Robust Digital Image Watermarking Method using Wavelet-Based Fusion,” in Proc. Int. Conference on Image Processing, Vol.1, Oct. 1997,.544-547.

[10] F.M. Boland, J.J.K. 0 Ruanaitdh and C. Dautzenberg, “Watermarking Digital Images for Copyright Protection,” International Conference on Image Processing And its Applications, Fifth International Conference, 4-6 July 1995,326-330 .

[11] Chin-Chen Chang and Hsien-Chu Wu, “Computing Watermarks from Images Using Quadtrees,” in Proceedings of the IEEE, 2000,.123-128,.

[12] Chuhong Fei, Deepa Kundur and Raymond KwongEdward S.Rogers, “The choise of watermark domain in the presence of compression,” in Proceedings of the IEEE, 2001,79-84.

[13] Patrizio Campisi, Deepa Kundur, “Robust Digital Watermarking in the Ridgelet Domain”, IEEE Signal Processing Letters, Vol.11,No.10,Oct-2004,.826-830.

[14] Chin-Shiuh Shieh,Hsiang-Cheh Huang,Feng-Hsing Wang and Jeng-Shyang Pan, “An Embedding Algorithm for Multiple Watermarks”, [15 ] juan r. hern andez, fernando p erez-gonz alez,“ Statistical Analysis of Watermarking Schemes for Copyright Protection of Images,” proceedings of the ieee, vol. 87, no. 7, july 1999,1142-1166,

[16]Saraju P. Mohanty, Renuka Kumara C., and Sridhara Nayak, FPGA Based Implementation of an Invisible-Robust Image Watermarking Encoder G. Das and V.P. Gulati (Eds.): Springer-Verlag Berlin Heidelberg 2004(CIT) 2004, LNCS 3356, 2004,344–353.

[17]R. Chandramoulia, Benjamin M. Graubardb, and Colin R. Richmondb,“A Multiple Description Framework for Oblivious Watermarking,” In Proc. of Security and Watermarking of Multimedia Contents III, SPIE vol. 4314, 2001.

[18]Chuhong Fei, Deepa Kundur, Raymond, H.Kwong,Analysis and design of secure watermark-based authentication systems, IEEE transactions on information forensics and security,Vol.a,No,1,March 2006,43-55.

[19] Manuel Alvarez-Rodriguez,Fernando Perez-Gonzalez,Analysis of piolet-based synchronization algorithms for watermarking of still images,SSiginal communication image communication ,17,2002,611-633.

[20] Deepa Kundur,Dimitrios Hatzinakos,Towards robust logo watermarking using multiresolution image fusion principles,IEEE transactions on multimedia,Vol.6,No.1,Feb.2004,185-198.

Attached Files:

• dip01128.doc (195 KB)