signalspot.com

 

ABSTRACT

Digital Watermarking and data hiding have gained popularity in recent years as a means of protecting digital images from theft, illegal copying and unlawful reproduction. Some digital watermarking algorithms were proposed using spatial domain and transform domain techniques. The transform domain could be DCT, DFT or DWT. Several digital watermarking algorithms using wavelet transform are available in the literature. In this paper a watermarking algorithm which is non-blind (requires the presence of host image for detection), and robust is proposed. The proposed algorithm is more robust to some attacks like median filtering, row-column copying, and row-column blanking, not discussed in the method proposed by Ganic and Eskicioglu [1]. The proposed watermarking scheme embeds the watermark in the Discrete Wavelet Transform domain. The watermark is a binary logo and is embedded in the approximate subband of the wavelet decomposition. The basis for embedding is the contrast of the each individual nxn block in the LL subband of wavelet domain. The pixel intensities of each individual block is modified within the range specified by the contrast value of the block. The watermarking scheme is robust to several attacks like median filtering, cropping, rotation, scaling, and compression, high pass filtering low pass filtering, row-column blanking, row-column copying and salt & pepper noise.

(Note: This Paper was presented in ICSIP., Signalspot Please Download the paper

for proper formatting, images,equations and symbols)

 

KEY WORDS

Digital Image Watermarking, Discrete Wavelet Transform Haar Wavelet.

1 Introduction

Over the past few years digital watermarking has become popular due to its significance in content authentication and legal ownership for digital multimedia data. A digital watermark is a sequence of information containing the owner’s copyright for the multimedia data [11]. It is inserted invisibly in another image (host image) so that it can be extracted at later times for the evidence of rightful ownership. Available digital watermarking techniques can be categorized into one of the two domains, viz., spatial and transform, according to the embedding domain of the host image. The simplest technique in the spatial domain algorithms is to insert the watermark image pixels in the least significant bits (LSB) of the host image pixels [8, 7]. The data hiding capacity in these algorithms is high. However, these algorithms are hardly robust for various attacks and prone to tamper by unauthorized users. Watermarking in transform domain is more secure and robust to various attacks. The wavelet transform decomposes the image into four subbands called LL, LH, HL and HH. Wavelet transform is computationally efficient and it reflects the anisotropic properties of HVS. Magnitudes of DWT coefficients are larger for LL band compared to other bands. The larger the coefficient the more significant it is. Using DWT an image can be shown at different levels of resolution. The blocking artifact problem is less severe in DWT than DCT, since there is no block processing in DWT. A wavelet transform provides both frequency and spatial description for an image. Wavelet based watermarking techniques are getting popularity because of JPEG2000 standard.

1. 1. Related Work

 

Wavelet based watermarking algorithms are categorised based on the availability of secret key and host image in the process of detection. They are blind, semi-blind and non-blind watermarking techniques. Xia et al. [12] presented a watermarking scheme which is a non blind in nature. They embedded in all the bands except LL band. Hsu and Wu [8] proposed a watermarking scheme wherein the wavelet transform is applied to both the host image and binary watermark . Raval and Rege [3] presented a multiple watermarking scheme. Here multiple watermarks are embedded in the low frequency and high frequency bands of DWT so that the watermark is robust to various attacks.

 

An additive watermarking technique using image fusion technique is proposed by Kundur and Hatzinakos [4]. A multiple watermarking scheme which uses all the bands is proposed by Tao and Eskicioglu [5]. Different scaling factors were used for different bands. A combination of SVD and DWT is used by Ganic and Eskicioglu [1]. All the above said algorithms are non-blind i.e., they require the original host image to extract watermark. This is a serious limitation of the algorithms. Tsai [9] presented a blind watermarking algorithm based on scalar quantization. However it is not robust to many geometric attacks. A variable quantization algorithm was proposed by Chen at al [13]. The algorithm proposed by Chen at.al., is robust to many attacks like scaling, resizing , cropping etc., However it is not robust against attacks like rotation.

2. Proposed Scheme

 

A Discrete wavelet transform when applied to an image transforms the image to images of various resolutions. A one level DWT decomposition gives four subbands., namely LL, HL,LH and HH. These subbands are shown in Fig 3. Most of the energy is contained in the LL band. The proposed watermarking scheme embeds the watermark in the LL band using HAAR wavelet.

 

2.1 Watermark Embedding Algorithm

 

Step 1: Wavelet Transform

 

A one level DWT decomposition of the original host image is obtained. Four bands LL, HL, LH and HH are the host images of various resolutions.

 

Step 2: The approximate subband LL is selected for embedding. LL band is divided in to various blocks of size 4×4.

 

Step 3: Compute the mean of the block and classify each pixel x(i,j) into one of two categories. If the pixel’s intensity value is greater than the mean of the block it is XL category, else it is of XH category.

 

Step 4: Compute the means ML and MH for the two categories XL and XH.

 

Step 5: Contrast value of a block is defined [15]

 

C=max(Cmin, ?(xmax-xmin))

 

? is a constant and Cmin is a constant which defines the minimal value a pixel’s intensity can be modified.

 

Step 6: Modify the pixel intensities of a block as follows:

 

If the watermark bit is a ‘1’ then

xnew=xmax if x > MH

xnew=xmean if ML < x < xmean

xnew=x+µ

 

If the watermark bit is a ‘0’ then

xnew=xmin if x < ML

xnew=xmean if xmean < x < MH

xnew=x-µ

 

xnew is the new intensity value of the pixel, x is the original intensity value of the pixel, µ is a random value between 0 and C.

 

Step 7: The modified pixels are positioned in the original place.

 

The pixel intensity values are modified according to the contrast value of the block. If the contrast value is large then the pixels are modified more else if the contrast value is small the modification is less. So if a ‘1’ is embedded into a block then the average intensity value of the block is more than the original. If a ‘0’ is embedded then the average intensity value is less than the original host. An offset parameter µ is used for achieving additional robustness .

 

 

2.2 Watermark Extraction Algorithm

 

 

The extraction algorithm requires the original host image. DWT using HAAR wavelet is applied to the host image. From the LL subband the average intensity of each block is calculated and then compared with the watermarked image. If the average intensity of a block of watermarked image is greater than the average intensity of the same block of host image then a ‘1’ is extracted from the block. If the average intensity is less than the host image a ‘0’ is extracted.

2.3 Experimental Results and Observations

 

Experiments were conducted using Lena as host image. The two images host image and watermark image are shown in Fig 1 & 2 respectively. The size of the host image is 512×512. The size of the watermark image is 64×64. A Haar Wavelet filter is used for wavelet decomposition. A Haar wavelet transform is conceptually simple and fast. It is exactly reversible without any edge effects. The Host image is decomposed into four subbands LL,LH,HL & HH. This is shown in Fig 3. The watermark image is embedded in the LL subband. Out of all the subbands the subband LL is having the largest wavelet coefficients. Various attacks used to test the robustness of the watermark are JPEG compression, Rotation, Resizing, Lowpass filtering, Median filtering, cropping, row column blanking, row column copying, salt & pepper noise and highpass filtering. The visual appearance of the watermarked image is good showing no significant artefacts or distortions because of the process of watermarking.

 

Fig 3 1-Level Haar Wavelet Decomposition

 

 

The extracted watermarks after applying various attacks are shown in Fig 5. The watermarked image is compressed using lossy JPEG compression using MATLAB. The index of the JPEG compression ranges from 0 to 100, where 0 is best compression and 100 is best quality. The reconstructed watermarks for various indices are shown in Fig 5. The proposed scheme works well even for extreme compression.

The watermarked image is rotated by 10⁰,30⁰,90⁰, and 120⁰ to the right and then rotated back to their original position using bilinear interpolation. The recovered watermark is very good for all the rotations specified. Resizing operation first reduces or increases the size of the image and then generates the original image by using an interpolation technique. This operation is a lossy operation and hence the watermarked image also looses some watermark information. In this experiment first the watermarked image is reduced from 512×512 size to 256×256. By using bilinear interpolation its dimensions are increased to 512×512. The extracted watermark as shown in Fig 5 is good. For lowpass filtering attack a 3×3 mask consisting of 0.9 intensity values is used. The recovered image is distinguishable showing its resistance to lowpass filtering attack. Median filter is a non linear spatial filter which is usually used to remove noise spikes from an image. The watermarked image is attacked by median filtering with a 3×3 mask. The median filtered watermarked image is more blurred than low pass filtered image. The extracted watermark is visually recognizable. Cropping operation deletes some portion of the image. This is a lossy operation. In this experiment half of the watermarked image is cropped and then watermark is extracted. The extracted watermark is still recognizable even after 50% cropping . In Row Column blanking attack , a set of rows and columns are deleted. In this experiment 10,30,40,70,100,120 &140 rows and columns are removed. The extracted watermark is showing good similarity with the original watermark. In Row-Column Copy attack a set of rows and columns are copied to the adjacent or random locations. In this experiment 10^th row is copied to 30^th row, 40 to 70, 100 to 120 and 140 to 160. The extracted watermark is clearly visible. The watermarked image is attacked by salt & pepper noise with a noise density of 0.01. The extracted watermark even though is recognizable, is not good compared to the watermark extracted by various other attacks. Finally the proposed algorithm also shown resistance to high pass filtering operation as shown in Fig 5.

 

The Normalized Correlation value ‘NC’ defined in [14] is used as a metric to compare the robustness and summarized in Table 1.

 

Type of Attack	Characteristic of Attack	NC Value
JPEG Compression	Index-10	0.8786
	Index-20	0.8771
	Index-30	0.8759
	Index-40	0.8587
	Index-50	0.8629
	Index-60	0.8613
	Index-70	0.8523
	Index-100	0.8772
Rotation	100	0.8739
	300	0.8306
	900	0.9975
	1200	0.8321
Resizing	512-256-512	0.8345
Lowpass Filter	3×3 mask	0.8584
Median Filter	3×3 mask	0.8670
Cropping	50% (right)	0.7044
Row-Column Blanking	10,30,40,70,100, 120,140	0.8601
Row-Column Copying	10,30,40,70,100, 120,140,160	0.8619
Salt &Pepper Noise	0.01 noise density	0.5765
High-pass Filter	DFT filtering	0.8164

 

Table 1 NC values for Various Attacks

 

3. Conclusions

 

In this paper a non blind watermarking algorithm based on contrast of the block in LL band of the wavelet domain is presented. The watermark image is embedded in the LL sub-band. The proposed algorithm is shown to be robust to 9 attacks ., JPEG compression, rotation, scaling, cropping, median filtering, lowpass filtering, row-column copying, row-column blanking and salt &pepper noise. The proposed algorithms is robust to attacks like row-column blanking, row-column copying and median filtering, not discussed in [1]. This means that an embedded watermark is still recoverable even after common image processing operations on the watermarked image. Non-blind nature of the proposed algorithm is a serious limitation.

 

References

 

[1] E. Ganic and A. M. Eskicioglu, “Robust digital watermarking: Robust DWT-SVD domain image watermarking: embedding data in all frequencies,” presented at Proceedings of the 2004 Multimedia and Security Workshop on Multimedia and Security, 2004.

 

[2] D. Kundur and D. Hatzinakos, “Towards Robust Logo Watermarking using Multiresolution Image Fusion,” IEEE

Transactions on Multimedia, vol. 6, pp. 185-198, 2004.

 

[3] M. S. Raval and P. P. Rege, “Discrete wavelet transform based multiple watermarking scheme,” presented at Convergent Technologies for the Asia-Pacific Region, Bangalore, India, 2003.

 

[4] D. Kundur and D. Hatzinakos, “Towards Robust Logo Watermarking using Multiresolution Image Fusion,” IEEE

Transactions on Multimedia, vol. 6, pp. 185-198, 2004.

 

[5] P. Tao and A. M. Eskicioglu, “A Robust Multiple Watermarking Scheme in the Discrete Wavelet Transform Domain,” presented at Symposium on Internet Multimedia Management Systems V, Philadelphia, PA, 2004.

 

[6] X. Xia, C. G. Boncelet, and G. R. Arce, “A multiresolution watermark for digital images,” presented at Proceedings 4th IEEE International Conference on Image Processing, Santa Barbara, CA, 1997.

 

[7] C. S. Lu, S.-K. Huang, C.-J. Sze, and H.-Y. Liao, “A new watermarking technique for multimedia protection,” presented at Multimedia Image and Video Processing, Boca Raton, FL, 2001.

 

[8] C. T. Hsu and J. L. Wu, “Hidden Digital Watermarks in Images,” IEEE Transactions on Image Processing, vol. 8, pp. 58-68, 1999.

 

[9] M. J. Tsai, K. Y. Yu, and Y. Z. Chen, “Joint Wavelet and spatial transformation for digital watermarking, ” IEEE

Transactions on Consumer Electronics, vol. 46, pp. 241-245, 2000.

 

[10] A. Lumini and D. Maio, “A Wavelet-Based Image Watermarking Scheme,” presented at The International Conference on Information Technology: Coding and Computing, (ITCC’00), Las Vegas, Nevada, 2000.

 

[11] J. Huang, Y. Q. Shi, and Y. Shi, “Embedding Image Watermarks in DC Components,” IEEE Transactions on Circuits and System for Video Technology, vol. 10, pp. 974-979, 2000.

 

[12] Xia, L., Boncelet, G., Acre, G.R., “Wavelet transform based watermarking for digital images”, in Optics Express, Vol.3, no.12, Dec 1998.

 

[13] Chen, T.Z., Horng, G., Wagng, S.-H., “ A Robust Wavelet Based Watermarking Scheme using Quantization and Human Visual System Model”, in Proceedings of the Pakistan Journal of Information and Technology, vol.2., no.3., pp.212-230, 2003.

 

[14]Frank Y. Shih, S.Y.T.Wu., “Combinational image watermarking in spatial and frequency domains” Pattern Recognition 36 (2003) 969-975.

 

[15] G.S.Gulstad, K.Bruvold., “An Adaptive Digital Image Watermarking Technique for Copyright Protection” A report.

(Note: This Paper was presented in ICSIP., Signalspot Please Download the paper

for proper formatting, images,equations and symbols)

Attached Files:

• dip013651.doc (557 KB)