Abstract -A robust watermark scheme for copyright protection is proposed in the present paper. The present method selects the pixel locations to insert the watermark by a new technique called fuzzy based wavelet approach. The watermark is embedded in the sorted pixel locations of fuzzy based wavelet approach by using pixel value difference method. The proposed approach overcomes the weak robustness problem of embedding the watermark in the spatial domain and also in pixel value difference method. Further the watermark extraction does not require the original image as in the case of many digital watermarking methods. The fuzzy logic approach in the wavelet domain eliminates the requirement of repeated embedding process. The experimental results indicate the high image quality and robustness against various attacks when compared to several approaches.

Keywords : Fuzzy Approach, Digital Image Watermarking, Sorted Pixel Value Difference (SPVD).

n recent years, watermarking has become an attractive topic and many watermarking schemes have been proposed [3], [6]. Many literatures have reported about watermarking based on spatial domain with different conventional extraction techniques [4]. Watermarking in the frequency domain is more robust than watermarking in the spatial domain [1], because the watermark information can be spread out over the entire image [2]. Transform domain watermarking techniques are more robust, due to the fact that when image is inverse wavelet transformed, watermark is distributed irregularly over the image, making the attacker difficult to read or modify.

In 2002, Joo et al. proposed a robust watermark scheme by embedding a watermark into wavelet low frequency sub-band [5]. According to the embedding location, the watermark is extracted by comparing the two wavelet coefficients by three level wavelet transforms LL3 and LL3'. Finally, the extracted watermark is compared with the original watermark by similarity measure formula. Although the above scheme provides the characteristics of robustness and imperceptibility, but the embedding process is quite time-consuming. Besides, the original image is required in the watermark extraction process, which is impractical in real application. In one of the recent method [7], the original image is transformed into wavelet coefficients by one-level wavelet transform first. Then, three highfrequency sub-bands are modified and obtain its reference image by performing inverse wavelet transform. The watermark is embedded into the reference values between the original image and its reference image. In the watermark extraction process, the watermark extraction does not require the original image. This method suffers from poor identification of pixel locations where watermark is to be inserted. To overcome the above disadvantage, the proposed method initially locates the pixel locations by applying fuzzy logic on two level wavelet coefficients and further they are sorted to insert the watermark by PVD method [8]. This makes the present method more powerful then earlier methods in breaking the watermark and also it gives high robustness.

The rest of this paper is organized as follows: The section 2 presents a brief introduction of the PVD; Section 3 describes the proposed method for embedding and extraction of watermark. The section four describes the experimental results. The conclusion is discussed in the final section.

The basic PVD scheme [8] is meant for steganographic images. This scheme offers high imperceptibility to the stego image by selecting two consecutive pixels as the object of embedding. The payload is determined by the difference value between the pixels as given below. The basic PVD method, determines whether the two consecutive pixels belong to an edge or smooth area by checking out the difference value between two consecutive pixels. If the difference value is large, i.e. the two pixels are located in an edge area; more secret data can be hidden here. On the contrary, if the difference value is small, i.e. the two pixels are located in a smooth area, less secret data can be embedded. Therefore, this scheme produces stego images that are more similar to the original images than those produced by LSB substitution schemes, which directly embed secret data into the cover image without considering the differences between adjacent pixels.

Given a cover image I of size M×N, I i is a sub block of I that has two consecutive pixels broken down by partitioning I in raster scan order such that I={ I i /i=1,2,??..(M×N)/2}. By definition each I i has two elements E (r,L) and E (r,R) . The pixel values of E (r,L) and E (r,R) are V (r,x) and V (r,y) respectively. The difference value d i of V (r,x) and V (r,y) can be derived by Equation (1).
d i = |V(r,x) -V(r,y) |(1)
The range Table 1 which consists of n contiguous sub ranges T j ; T= {T j | j=1, 2,..n}, provides major information about the hiding capacity of Ii. Each sub range T j has its lower and upper bound values, say l j and uj, so that it will have T j ? [l j , u j ]. The width w j of each T j is selected to be a power of 2, and can be computed by w j =u j -l j +1.Each sub block I i relates to its sub range T j from the range Table 6.1 such that T j =min(d i ,w j ) and d i ?[l j ,u j ]. This way, the hiding capacity of two consecutive pixels can be obtained by Equation (2). bi = log (w j )

Here, b i is the number of bits that can be hidden in I i . Table 1 clearly indicates the number of bits of watermark to be selected. The PVD method selects bi bits from the binary secret data stream and transform bi into its decimal equivalent value bi'. Then it computes the remainder values E rem(r,x) , E rem(r,y) and I rem(r) of E (r,x), E (r,y) and sub block I i respectively by using the following Equations ( 3).
E rem(r,x ) = E (r,y) mod 2 bi E rem(r,y) =E (r,y) mod2 bi(3)I rem(r) = (E (r,x) + E (r,y) ) mod 2 bi
After this the PVD embeds b i bits of secret data into I i by altering E (r,x) and E (r,y) . The optimal approach for altering the E (r,x) and E (r,y) to achieve the minimum distortion is as follows:
Case1: I rem(r) > b i' , m?(2 bi )/2 , E (r,x) ) ? E (r,y) (E' (r,x) ,E' (r,y) )=(E (r,x) -m/2 ,E(r,y)-m/2 2) Case 2: I rem(r) > b i ', m<=(2 bi )/2, E (r,x) <E (r,y) (E' (r,x), E' (r,y) )=(E (r,x) -m/2 ,E (r,y) -m/2 2) Case 3: I rem(r) > b i ', m>(2 bi )/2, E (r,x) >=E (r,y) (E' (r,x), E (r,y) )=(E (r,x) + m 1 /2 , E (r,y) + m 1 /2 2) Case 4: I rem(r) > b i ', m>(2 bi )/2, E (r,x) <E (r,y) (E' (r,x) ,E'(r,y))=(E(r,x)+ m 1 /2 , E (r,y) + m 1 /2 2) Case 5: I rem(r) > b i ', m>(2 bi )/2, E (r,x) ?E (r,y) (E' (r,x), E' (r,y) )=(E (r,x) + m/2 2, E (r,y) + m/2 2) Case 6: I rem(r) ? b i ', m>(2 bi )/2, E (r,x) <E (r,y) (E (r,x) ,E' (r,y) )=(E (r,x) + m/2 2, E (r,y) + m/2 2) Case 7: I rem(r) ? b i' , m>(2 bi )/2, E (r,x) ?E (r,y) (E' (r,x) , E' (r,y) )= (E (r,x) -( m 1 /2 , E (r,y) -m 1 /2 2) Case 8: I rem(r) ? b i' , m>(2 bi )/2, E (r,x) ?E (r,y) (E' (r,x), E' (r,y))=( E (r,x) -m 1 /2 , E (r,y) -m 1 /2 2)
In the above approach, m=|I rem(r) -b i ' |, m 1 =2 bi -|I rem(r) -b i ' | and E' (r,x) , E' (r,y) are new pixels values after embedding bi bits of the secret data into sub block I i .

In the recovery process, the secret data is extracted quickly without using the original image. It is essential to use original range Table 1 designed in the embedding phase in order to figure out the embedding capacity for each sub block I i . Given a sub block I i with two consecutive pixels from the watermarked image with their pixel values being E (r,x) and E (r,y) respectively, the difference value di of E (r,x) and E (r,y) can be derived by Equation (1). Each I i can be related to its optimal sub range T j from the original Table 1 according to the difference value d i . Hence, the width of the sub range by w j =u j -l j +1is computed and the number of bits bi of secret data can be extracted from I i by Equation (1). The value of b i ' is computed by using the Equation (4). b i '=(E' (r,x) +E' (r,y) )mod2 bi (4)

Then transform b i ' value into a binary string with the length b i . If di value is zero or either or both of E' (r,x) , E' (r,y) overflows the boundary 0 or 255, consider the three situations below where the falling off boundary problem happens and revise the binary string with the length b i .

Case 1: if E (r,x) ? 0 , E (r,y) ? 0 and E' (r,x) < 0 or E' (r,y) < 0 , then re adjust E' (r,x) and E' (r,y) to be E" (r,x) and E" (r,y) by (E" (r,x) ,E" (r,y) ) = (E' (r,x) + (2 bi ) / 2 ), E' (r,y) + (2 bi ) / 2 )


# Global Journal of Computer Science and Technology Volume XII Issue IV Version I

After that, the recovery process is accomplished.

The proposed watermark embedding scheme contains three basic steps. The block diagram of FWSPVD is given in the Figure 1. The proposed FW method modifies the original image into transform domain and selects the pixel locations to insert a watermark in the difference values between the original image and its reference image based on a novel fuzzy logic in step one.

Step 1: In the first step, a novel FW approach is determined based on the pixel locations where watermark is embedded. DWT decomposes an image into subbands having a bandwidth approximately equal on a logarithmic scale. To achieve imperceptibility, the lowest band of the image is left unmodified. The gray level image is transformed into a DWT of both vertical and horizontal directions, resulting in one low frequency subband (LL) and three higher frequency subbands (LH, HL and HH). The same is repeated on LL subband to generate the next level of decomposition. This process can be repeated to n level decomposition by considering the length of watermark, robustness, fidelity and so on. The determined LL n can be seen as a reduced version of the original image. Based on this a reference LLn' is prepared by inverse wavelet transforming the original LL n by initializing the three high frequency subbands (LH n+1 , HL n+1 and HH n+1 ) excluding LL n+1 as zeros. The proposed FW approach is not selecting all those pixels that have the difference in LL n ' and LL n . The difference between LL n and LL n ' mainly ranges from -1 to +1, because the error content in the wavelet transform is minimum, that is the reason one always obtains the original image by inverse transformation. In the proposed approach the fuzzy difference between LL n and LL n ' is obtained for selecting the pixel locations where watermark is embedded. The pixel locations to embed watermark are identified by taking the difference between (LL 2 ?LL 2 ') based on FW approach. The proposed FW approach divides the range -1 to +1 in to four regions as R0, R1, R2 and R3 as shown in the Figure 2. The pixel locations are selected based on the FW algorithm.

The pixel locations are selected for the embedding of watermark if they fall in the fuzzy region R1 and R2. Finally, the watermark information is embedded into the subband LL 2 .


# Fig.2. Representation based on fuzzy wavelet approach


# F ebruary

Fuzzy Wavelet algorithm: begin for i =1 to n for j = 1 to m f(i,j)=LL n (i,j) -LL n '(i,j) if ((f(i,j) < 0.5) and (f(i,j)> -0.5))

then P(i,j) is considered for inserting the watermark else P i (i,j) is not considered for inserting the watermark end

Step 2 : The second step, groups only those pixel coordinates selected by step one based on their pixel values in ascending order. If two or more pixels are having same values then they will be sorted by row wise positions.

Step 3 : The pixel value differencing method only inserts watermarking bits in the adjacent pixels location based on pixel differences. By this it is easy to break or detect the watermark. To overcome this, the proposed method initially selects the pixel location where watermark is to be inserted based on FW approach. The watermark is inserted on the group of pixels in the sorted order by using SPVD method in the second level of wavelet transformed image as indicated in step two. By inserting watermark, the pixel value may be changed; and the sorted order may also be changed. The quality of the watermark is affected if the sorted order is changed after inserting the watermark bits by pixel value differencing method. To overcome this, in the proposed SPVD method watermark is inserted in the group of two pixels, if its values after inserting watermark are less than the next group of values.


# b) Watermark extraction process

Transform the watermarked image into wavelet coefficients by second level wavelet transformations. To extract the watermark signal, the sequences of embedding locations are utilized. Perform the inverse FWSPVD scheme to obtain the pixel locations and watermark contents.

Eight 256 × 256 sized cover images are used in the following experiments. As shown in Figure 3, those are Lena, Baboon, Pepper, House, Barbara, Milkdrop, F16 and Boat. The watermark considered for the experiments is logo SRRF GIET of size 32×32 as shown in Figure 4. Table 2 shows the PNSR and NCC values for all the cover images. From the Table 2 it is clearly evident that all the images shows high PSNR and NCC values which indicates robustness and high quality of image after watermark insertion.  


# F ebruary

The proposed method is tested with various attacks and the quality parameters are listed in tables. Table 3 shows the PSNR and NCC values with various attacks using the proposed FWSPVD method on the watermark images with SRRF GIET respectively. The PSNR and NCC values of Table 3 clearly indicate the robustness and quality of the image is not degraded for all attacks.

The resultant image after adding Gaussian (10%), Salt and Pepper (10%), and Poisson noise (10%) to the watermarked image by the proposed FWSPVD is as shown in Figure 5, which demonstrates that FWSPVD scheme is significantly robust against these noises. Median filtering with different window sizes is applied to the watermarked images of the FWSPVD method and the resultant images are shown in Figure 6, which reflects the maximum detector response. Figure 7 is the Gaussian blurred watermarked image by the FWSPVD method which demonstrates that FWSPVD scheme is significantly robust against this noise. Figure 8 shows, the results after rotating the watermarked image by 2, 3 and 4 degrees in order to keep the same size as the original image by which four corners of the rotated image are cropped. The extraction algorithm extracted more than 50% of the watermark.  4 compares the PSNR values after inserting the watermark without attacks by the proposed FWSPVD method with various other methods [7,8,9]. Table 4 clearly indicates the FWSPVD outperforms the other existing methods. A graph is also plotted in Figure 9 which indicates the comparison of the proposed FWSPVD method with various other methods without attacks.  By transforming the original image in wavelet domain and embedding a watermark in the difference values based on fuzzy approach between the original image and its reference image, the proposed scheme overcomes the weak robustness problem of embedding watermark in the spatial domain. Our approach does not require the original image for watermark extraction. The experimental results on various images with various attacks show that the proposed technique provides good image quality and robustness when compared to other methods. This factor is clearly evident from the table 2, table 3 and table 4.

The authors would like to express their gratitude to Sri K.V.V. Satyanarayana Raju, Chairman, and Sri K. Sasi Kiran Varma, Managing Director, Chaitanya group of Institutions for providing necessary Infrastructure. Authors would like to thank the anonymous reviewers for their valuable comments. 
1![Fig.1. Block diagram of the proposed FWSPVD method a) Watermark embedding process](image-2.png "Fig. 1 .")
3![Fig.3 Cover images (a) Lena (b) Baboon (c) Pepper (d) House (e) Barbara(f) Milkdrop (g) F16 (h) Boat](image-3.png "Fig. 3")
5678![Fig. 5 Watermarked image corrupted by different noises (a) Gaussian noise (10%) (b) Salt and pepper noise (10%) (c) Poisson noise (10%)](image-4.png "Fig. 5 Fig. 6 Fig. 7 Fig. 8")
9![Fig. 9 Performance comparison of the proposed FWSPVD method with existing schemes](image-5.png "Fig. 9")

1Rangel ju jw jb i=log(w j )T1=[0,7]0783T2=[8,15]81583T3=[16,31]1631164T4=[32,63]3263325T5=[64,127]64127646T6=[128,255]1282551287
2ImagePSNRNCCLena44.810.96Baboon46.790.97Pepper45.450.99House47.280.98Barbara44.700.96Milkdrop47.640.97F1649.810.97Boat47.370.97Global Journal of Computer Science and Technology Volume XII Issue IV Version I 44 2012 © 2012 Global Journals Inc. (US)
3201245 F ebruary
4Test imagesJiang-Lung Liu MethodG.Thirug nanam MethodMethod Chung Ming WangProposed FWSPVDPSNR(dB)Lena34.8739.9844.144.812012Baboon Pepper32.14 31.1134.45 36.5640.3 43.346.79 45.45House30.4934.9543.547.28Barbara33.1541.6242.544.70Milk drop32.6739.1445.947.6446F1633.7241.1543.549.81Boat31.2440.3242.147.37© 2012 Global Journals Inc. (US)
			© 2012 Global Journals Inc. (US) Global Journal of Computer Science and Technology Volume XII Issue IV Version I 43
		
		
			

				


* 
	
		
		
			SanghyunJoo
		
		
			YounghoSuh
		
		
			JaehoShin
		
		
			HisakazuKikuchi
		
	
	
		?A new Robust Watermark Embedding into Wavelet DC Components?,ETRI Journal
		
			24
			October 2002
		
	




* 
	
		
			VPotdar
		
		
			SHan
		
		
			EChang
		
		? A Survey of Digital Image Watermarking Techniques,? in Proc. of the IEEE International Conference on Industrial Informatics
				Perth, Australia
		
			2005
			
		
	




* 
	
		A Robust watermarking scheme using self reference image
		
			Jiang Lung Liu
		
		
			Der Chyuan Lou
		
		
			ChangMing
		
		
			Chang
		
		
			Hao Kuan
		
		
			Tso
		
	
	
		computer standards and interfaces
				
			Elsevier
			2006
			
		
	




* 
	
		A high quality steganographic method with pixel-value differencing and modulus function
		
			Chung-MingWang A, Nan
		
		
			-IWu A
		
		
			Chwei-ShyongTsai B
		
		
			Min-ShiangHwang
		
	
	
		The Journal of Systems and Software
		
			2007
		
	




* 
	
		
		
			GThirugnanam
		
		
			SArulselvi
		
	
	
		Robust Digital Image Watermarking Scheme Based on DWT and ICA? Global Journal of Computer Science and Technology
		
			10
			11
			October 2010
		
	




* 
	
		Techniques for data hiding
		
			WBender
		
		
			DGruhl
		
		
			NMorimoto
		
		
			ALu
		
	
	
		IBM Systems Journal
		
			35
			3&4
			
			1996
		
	




* 
	
		Secure spread spectrum for multimedia
		
			I JCox
		
		
			JKillian
		
		
			TLeighton
		
		
			TShammon
		
	
	
		IEEE Transactions on Image Processing
		
			6
			12
			
			1997
		
	




* 
	
		Seyed, -A Robust image watermarking in the ridgelet domain using universally optimum decoder
		
			KNima
		
		
			M
		
	
	
		? IEEE Transactions on Circuits and Systems for Video Technology
				
			n0.3, 2010
			20
			
		
	




* 
	
		
			JIngemar
		
		
			Cox
		
		
			LMatthew
		
		
			JeffreyAMiller
		
		
			Bloom
		
		Digital Watemarking and