Multimed Tools Appl DOI 10.1007/s11042-016-3902-4

A robust blind hybrid image watermarking scheme in RDWT-DCT domain using Arnold scrambling Soumitra Roy 1

& Arup Kumar Pal

2

Received: 18 April 2016 / Revised: 15 August 2016 / Accepted: 24 August 2016 # Springer Science+Business Media New York 2016

Abstract To compromise between imperceptibility and robustness property of robust image watermarking technique, a RDWT-DCT based blind image watermarking scheme using Arnold scrambling is presented in this paper. Firstly, RDWT (Redundant Discrete Wavelet Transform) is applied to each gray scale cover image block after the image is decomposed into fixed size non overlapping blocks. Secondly, the binary watermark logo is encrypted by Arnold chaotic map and reshaped to a sequence to improve the security of the logo. In the subsequent step, DCT (Discrete Cosine Transform) is employed on each LH subband of the non-overlapping host image block. Finally, after zigzag scanning of each DCT block a binary bit of watermark is embedded into each block by adjusting some middle significant AC coefficients using repetition code. Experimental results show that robustness is achieved by recovering satisfactory watermark data from the reconstructed cover image after applying common geometric transformation attacks (like rotation, cropping, scaling, shearing and deletion of lines or column operation etc.), common enhancement technique attacks (like lowpass filtering, histogram equalization, sharpening, gamma correction, noise addition etc.) and JPEG compression attacks. The proposed scheme is also tested to verify the robustness performance against standard benchmark software BCheckmark^ and satisfactory results are achieved against the Checkmark attacks such as Hard and Soft Thresholding, Template Removal, Warping, Dithering, Remodulation and Downsampling/Upsampling etc. Keywords Robustness . Blind watermarking . Redundant discrete wavelet transform . Discrete cosine transform . Repetition code . Arnold scrambling

* Soumitra Roy [email protected] Arup Kumar Pal [email protected]

1

Department of Computer Science and Engineering, Dr. B. C. Roy Engineering College, Durgapur, West Bengal 713206, India

2

Department of Computer Science and Engineering, Indian School of Mines, Dhanbad, Jharkhand 826004, India

Multimed Tools Appl

1 Introduction With the fast expansion of the world wide web and the rapid evolution of personal computer and digital data (including images, audio and video), access, transmit, save and distribution of digital data is become simple and very much time and cost effective over the internet. But in today’s forgery world, modification and perfect replication of digital data with the help of easyto-use software tools and cheap digital materials is turning out to be common practice for illegal uses while transmitting data over the Internet. Reason behind this security flaw in network technology is that there is no default safety mechanism for transmitting data. In general, security is conveyed as the form of confidentiality, integrity and authenticity/copyright protection. Confidentiality deals against unauthorised access of the digital data from illegitimate users. The aim of integrity property is to find or locate the modification of content by illegal users while authentication forbids the unlawful distribution of copyright contents. The confidentiality property has been preserved during transmission/preservation of digital data by cryptographic or a steganographic mechanism. Limitation of these types of techniques is that after the revelation of digital data there is no control against fraudulent manipulation or plagiarized redistribution of digital content. So the protection/security of digital data should be provided not only during their transmission time but for endurance as a precaution. In recent times digital watermarking has caught the considerable attention from modern scientists as an efficient and promising solution for this permanent security mechanism of digital data. Digital watermarking is the process of embedding identification code/watermark (especially a logo in form of the owner’s signature or company’s logo) which can be extracted later for copyright protection, ownership verification and content authentication. To enhance security and better effectiveness, and the basic required properties of the watermarking system are imperceptibility, robustness and embedding capacity. Imperceptibility means impossible to differentiate between original cover data and watermarked data by the human visual system (HVS). Existence of watermark data after manipulation of watermarked data to alter or destroy the authentication of cover data is known as the robustness of watermarking scheme. Amount of data embedded as watermark to extract the watermark effectively in receiver side is termed as watermark capacity. By increasing the embedded watermarking data capacity enhanced the watermarking scheme robustness which may affect the imperceptibility of the watermarked data. So there need a negotiation between these three properties to design an efficient watermarking scheme. In addition to these watermarking schemes should be readily embeddable and extractable for watermark data. Depending on the various properties, watermarking technique can be divided in various ways. According to the human perception, watermarking can be divided into visible and nonvisible watermarking. In the visible watermarking, transparent watermark is present in the cover image while preservation of the visual quality of watermark object and imperceptibility of watermark logo (by human perception) is the main concerned in the non-visible watermarking technique. Watermarking schemes can also be divided into three schemes as blind, semi-blind and non-blind watermarking [31] schemes. In the blind watermarking scheme neither original data nor watermark data are needed in detection process; where in a semi-blind scheme watermark data and some other side information is needed in detection process; while in non-blind watermarking scheme original data and sometime watermark data is needed in detection procedure. Depending on its application, watermarking schemes can be classified into robust watermarking schemes and fragile watermarking schemes. Robust digital watermarking techniques are applied for copyright protection. In robust digital watermarking

Multimed Tools Appl

technique embedded watermark information almost preserve the image perceptual quality while resisting all types of attacks or change to the watermarked image. Fragile watermarking schemes are widely used for data authentication and tamper detection. In fragile watermarking schemes, the aim is to embed a watermark that should be alterable under common signal processing attacks. A semi-fragile watermarking scheme which is the designed according to the integrity property of cover object is the mixture of robust and fragile watermarking scheme. Similar the robust watermarking technique, semi-fragile watermarking can bear some kind of modification to the watermarked object of legitimate distortion while like fragile one it can detect the region of the object that is modified by illegitimate distortion and distinguish them from the authentic or unaltered region.

1.1 Related work Recently, modern scientists designed many efficient watermarking schemes to satisfy the properties of digital image watermarking techniques. It starts from spatial domain watermark embedding schemes [6, 21, 27, 30, 31, 48], where watermark data is inserted by changing the pixel value of the gray level host image. Modern research suggests that spatial domain watermarks can be manipulated by some geometric attacks, so there exist serious drawback for the spatial domain watermarking scheme. But current study found that in transformed domain watermarking scheme where watermark information is embedded asymmetrically into the coefficients of a transformed image is not easily traceable and modifiable. So lots of watermarking scheme is developed in transform domain. Among these, A DCT domain robust image watermarking is proposed by Barni et al. [3] where random sequence is embedded into the selective DCT coefficients. Chu et al. [8] embeds the watermark into DCT transformed sub images which are constructed from sub sampling of the image. In [22] Liang et al. describes DCT based watermarking technique with lifting scheme. Deng et al. in [12] proposes a novel image watermarking algorithm where watermark binary image is embedded in the DC component of the non-overlapping block’s DCT coefficients. Das et al. in [11] designs DCT based watermarking scheme using correlation between DCT coefficients in the same position of adjacent blocks. A self-reference of specific coefficients in DCT domain based blind image watermarking is described in [53] where non-overlapping host image blocks are converted to DCT domain at first. Then DC coefficients of each block are used to predict AC coefficient in the central block and these estimated AC coefficients magnitude are modified according the watermark bit value B0^ or B1^. These AC coefficient prediction based [50] blind image watermarking suffers from insufficient estimation accuracy. Campisi et al. [5] and Liang et al. [23] designed their watermarking scheme based on Finite Ridgelet Transform (FRIT). Wavelet based digital watermarking is proposed by Kundur and Hatzinakos in [18]. In [49], Tay and Havlicek et al. discusses one wavelet based image watermarking technique where a scaled image is used as the watermark and embedded in the mid-frequency wavelet channel. In [40] Raval and Rege describes a DWT (Discrete Wavelet Transform) based multiple watermarking schemes where the cover image is decomposed in two levels and watermarks were inserted in LL (low frequency) and HH (high frequency) bands. In [55] a pseudorandom sequence is weighted using a weighting function and then embedded to the wavelet transform coefficients of a host image. A robust image watermarking is proposed in [33] using DWT and nonnegative matrix factorization (NMF). A Discrete Fourier Transform (DFT) based watermarking is described in [32]. A blind DFT based novel image watermarking is described in [37]. A digital watermarking technique where a circular symmetry based watermark is embedded in image

Multimed Tools Appl

using DFT domain is described in [47]. A DFT based watermarking scheme which can survive rotation, scaling and translation is proposed in [26]. In [52], a wavelet based blind image watermarking scheme using two trees is presented where two trees is quantized with quantization index. Lien and Lin [24] explains a method using four trees where four trees are quantized according to the embedded watermark bit value. In [25], a maximum wavelet coefficient quantization based blind image watermarking is proposed. Recently, singular value decomposition (SVD) based transform domain watermarking scheme has been widely used as its attractive properties and [2] geometric features from an image is easily obtained by SVD mathematical techniques. Basically, in SVD based watermarking scheme singular values (SVs) of an image are modified to embed watermark or watermark SVs. In [9], Chung et al. discusses a vector quantization and SVD-based image hiding algorithm scheme for embedding the secret data into the SV matrix of the SVD. Liu and Tan in [28] proposes a pure SVD based blind watermarking scheme. A block-based image watermarking algorithm is discussed in [7], where elements in U matrix is modified after divided the image into several blocks. To improve robustness spread spectrum along with SVD is used in [4]. In this technique actually two watermarks are embedded, one with the spread spectrum technique and other with SVD. With the aim to increase invisibility and capacity of SVD based watermarking scheme, Chung et al. [10] proposes two notes where modify the coefficients in both U and V components after SVD transformation. For further improvement of method in [10], Fan et al. [13] modify only the first column of U and V matrices after SVD transformation. In [16], Ghazy et al. divides the cover image into non-overlapping blocks and then SVs of these blocks are used to embed watermark. A chaos based image watermarking is proposed in [56], where scrambled watermark is inserted into the quantized DCT coefficients. Ma et al. [29] devises a DCT based image watermarking where Arnold scrambled watermark bits are used to modify the quantized DCT coefficients. In [14], selected DCT coefficients are modulated by BCH encoded watermark bits. Phadikar et al. designs a watermarking system using QIM (quantization index modulation) and DCT in [36]. To achieve better robustness and high imperceptibility, recently researchers have developed their watermarking schemes using two or three transform domain techniques. These hybrid domain techniques give better results than their single counterpart. Singh et al. [42] presents a robust and imperceptible hybrid image watermarking method in DWT-DCT-SVD domain where the watermark is embedded on the SVs of the cover image DWT sub bands and their technique needs less SVD computation than other methods. A nonblind DCT-SVD based hybrid domain watermarking is presented in [38]. In this technique DCT is applied to the cover image at first and then coefficients are scanned in zigzag order. Then SVs of the cover image is modified with the SVs of DCT transformed visual watermark. Its main disadvantage lies in its computational value. For identity authentication purposes a robust hybrid multiple watermarking technique using fusion of DWT, DCT and SVD is proposed in [41] where multiple watermarks are embedded into the same image simultaneously. In this technique image watermark information is embedded in the S component of the cover image where the text watermark considered as EPR (Electronic Patient Record) is embedding at the second level of the D (diagonal sub-band) of the cover image after encryption. A DWT-SVD based watermarking scheme is proposed in [15]. After decomposing the cover image into four sub bands, SVs of each sub band is modified by SVs of watermark data in this technique. One secure spread-spectrum technique based multiple watermarking scheme is presented in [45] where pseudo-noise (PN) sequences are generated corresponding to each watermarking bit and embedding of these sequences is done column wise into the selected DWT coefficients in the subband. In this technique, text watermarks like patient

Multimed Tools Appl

identification represented in binary arrays using ASCII code and doctor’s signature or telemedicine centre name represented in binary image format into host digital radiological image for potential telemedicine applications. For better security, error correcting code (ECC) is applied to the ASCII representation of the text watermark before embedding. A DWT-SVD based secure multiple watermarking technique is proposed in [43] where encryption and ECC is applied to the ASCII representation of the text watermark before embedding. Another hybrid domain robust and imperceptible secure multiple watermarking technique based on DWT, DCT and SVD is develop in [44] where encryption is applied to enhance the security of the ASCII representation of the text watermark before embedding. In [34], Pandey et al. embed four different watermarks (i.e. Signature, index, caption and reference watermark) in form of image and text using fusion of DWT and SVD. In this method, Secure Hash Algorithm (SHA512) is used for generating hash corresponding to iris part of the cover digital eye image and this unique hash parameter is used for enhancing the security feature of the proposed watermarking technique. A hybrid watermarking scheme using RDWT and SVD is proposed in [20]. Rastegar et al. in [39] proposed hybrid domain watermarking technique using SVD and radon transform. The effects of different ECCs on the robustness and imperceptibility of discrete wavelet transform and singular value decomposition based dual watermarking scheme is investigated by Singh et al. in [46] where four different ECCs such as Hamming, the Bose, Ray-Chaudhuri, Hocquenghem (BCH), the Reed–Solomon and hybrid error correcting (BCH and repetition code) codes are considered for encoding of text watermark in order to achieve additional robustness for sensitive text data such as patient identification code. In their scheme, text and image watermarks are embedded into cover radiological image for their potential application in secure and compact medical data transmission.

1.2 Main contribution of work Literature survey suggests that the robustness of image watermarking scheme can be improved with the suitable hybrid domain technique and extra security can be provided with scrambled watermark. In the following section, reason behind the selection of RDWT and DCT for this blind image watermarking technique and some special features of this proposed method is illustrated in details:

&

&

DWT based watermarking methods are common because of its excellent spatio-frequency localization properties which help to find the areas in cover image to embed watermark imperceptibly. Still the reason behind the application of RDWT than DWT in this proposed method is that during downsampling process of DWT, potentially valuable information of image may be removed which are not important to reproduce the image. Because of this discarded coefficients, little bit shift/modification in the image reduce the possibility to recover the watermark from host image. That is DWT suffers from poor directional information due to DWT’s shift variant property. To eliminate this shift variant property, RDWT is used which is shift invariant. This matter is described in detail in RDWT preliminary section. One of the major shortcomings of block based DCT is the Bblocking artifact^ which appears at block boundaries due to coarse quantization of the coefficients. To avoid this visible artifacts drawback, proposed DCT based watermarking scheme is actually developed on using some middle band coefficients after zigzag scanning of each DCT block. As these middle bands coefficient pairs are less vulnerable on modification than low and high coefficient pairs.

Multimed Tools Appl

&

&

In any watermarking scheme, watermarks (image/text) are related information about the cover multimedia objects. Confidentiality of this important information should preserve still unauthorized users extract these special messages. To improve the security of watermark, in this proposed technique Arnold scrambling is applied on binary watermark to create chaotic watermark. Proposed blind image watermarking technique use basic ECC repetition code to preserve one watermark bit in every decomposed non-overlapping host image block. At first chaotic watermark bit is repeated Bt^ times. In the subsequent step, select Bt^-pairs of DCT coefficients from middle bands. Then to preserve watermark bit= 1 in each coefficient pair, check if 1st coefficient is greater than 2nd coefficient. If condition satisfied, keep the coefficients unchanged. Otherwise swap both these values. To preserve watermark bit= 0, 1st coefficient should be less than 2nd coefficient in each coefficient pair. In conventional ECC based watermarking schemes, redundant bits are added to the host image for error detection or error correction on the receiver side. So, implementation of proposed repetition code based watermarking scheme is less complex and computational complexity is also minimum than other ECC based watermarking methods.

After this introductory section, the rest of the paper is organised as follows. Preliminaries of the Arnold scrambling, RDWT, block-based DCT and repetition code are presented in Section 2. The proposed watermark embedding and extraction algorithms are described in Section 3. Experimental results are furnished in Section 4. Finally, conclusions are drawn in Section 5.

2 Preliminaries 2.1 Arnold scrambling In order to expand the robustness of the algorithm and provide extra security to the embedded watermark, Arnold scrambling is employed in the preprocessing step of the proposed method. The classical Arnold scrambling method jumbles up the pixel positions of the host image to generate a chaotic image and thus takes the responsibility to act as secondary encryption technique. Eventually, the watermark is shared out in all space of the host image as space locations of watermark pixels are disturbed by scrambling method. This muddled watermark cannot be recovered without proper information about the scrambling algorithm; even the attacker successfully extracts the watermark from cover image. Hence, scrambling transformation improves the security of the embedded watermark and increases the robustness of the proposed method. Two dimensional Arnold scrambling transformations is defined as follows:
0


i

1
0
¼

j

1




0 0 1



i

mod N; i; j; i ; j ¼ f0; 1; 2; …::; N−1g 2

j


ð1Þ

where i,j is the pixel coordinates of the original space: i′, j′ is the pixel coordinates after iterative computation scrambling, N is the size of the watermark. To restore the original watermark, the corresponding inverse transformation formula can be defined as:

0












i


¼
2 −1

i 0
þ
N
mod N
−1 1

j

N

j


ð2Þ

Multimed Tools Appl

2.2 RDWT Wavelet transform is used extensively in image processing because both spatial and frequency resolution is permitted during wavelet decomposition. The DWT uses lowpass and high-pass filters, H(z) and G(z), and downsampling (↓) to analyze a digital image at different frequency bands with different resolutions by decomposing the signal into coarse coefficients and detail coefficients. The coarse coefficients are generated by the lowpass filter H(z) and detail coefficients are fabricated by high-pass filter G(z) along with downsampling. As a result of this downsampling process, potentially valuable information of image may be removed which are not important to reproduce the image. Because of this discarded coefficients, little bit shift/ modification in the image reduce the possibility to recover the watermark from host image. To overcome DWT’s shift variant property, RDWT is proposed which is shift invariant. In case of RDWT [54], decimator/downsampling in DWT which eliminate redundant coefficients are removed. Expanding a image using lowpass filter H(z) and high-pass filter G(z) is referred as analysis filters whereas by synthesis filters image is returned back to its original form. Figure 1 shows two dimensional DWT using analysis and synthesis filters on an 8 × 8 image. At first, analysis filters are applied to the rows of image with downsampling. This constructs one coarse row coefficient and one detail row coefficient with size 8 × 4. In the next step, four different image sub-bands (LL,LH,HL,HH) with size 4 × 4 are generated after applying analysis filters to the column of coarse row coefficient and detail row coefficient. Rows and columns analyzed with a high pass filter are designated with an H. Likewise, rows and columns analyzed with a low pass filter are designated with an L. For example, LH subband is produced using a low pass filter on the rows and a high pass filter on the columns. To reconstruct the 8 × 8 image, synthesis filter is used. In the first step, synthesis filters are applied to the columns of all four sub-bands with upsampling (↑). In the following step, reconstructed image is formed with the application of synthesis filters and upsampling as shown in Fig. 1. Figure 2 explains two dimensional RDWT on same image size. Here all the steps for analysis and synthesis filters are same like DWT. But as upsampling and downsampling is not present in both the filters, output image size in each step is same as the previous step. This is demonstrated in Fig. 2. In this transformation; requirements of storage space are higher as redundant components are stored for next level and it is more over-complete representation than DWT.

2.3 Block-based DCT The DCT is one of the popular and widely used signal decomposition as well as compression techniques that transform a signal from spatial domain representation into a spectral representation with an inherent ability to exhibit excellent energy compaction for the signal or image. It

Fig. 1 Two dimensional DWT analysis and synthesis filter banks on 8 × 8 image

Multimed Tools Appl

Fig. 2 Two dimensional RDWT analysis and synthesis filter banks on 8 × 8 image

basically transforms the signal as a sum of sinusoids of varying magnitudes and frequencies. DCT transformation is used for transferring pixel values of the image from one domain to the other and after transformation this image consists of one DC coefficient and multiple AC coefficients. The most of the information of transformed image stored in few low level frequency coefficients. Those low level frequency coefficients exist in the upper top left corner of the image. DC coefficient is the average of the pixels of the image and AC coefficients contains also contains the significant information of the image but less than that of DC component. In block based DCT, the input image of size M × N is decomposed into nonoverlapping blocks of size m × n and then each block fb is transformed into corresponding DCT coefficients according to the following equation:     m−1 n−1 X X ð2x þ 1Þuπ ð2y þ 1Þvπ f b ðx; yÞcos F b ðu; vÞ ¼ αðuÞαðvÞ cos 2m 2n x¼0 y¼0

ð3Þ

where 8 rffiffiffiffiffiffiffiffiffiffi . > > < 1 m; αðuÞ ¼ rffiffiffiffiffiffiffiffiffiffi . > > : 2 m 8 rffiffiffiffiffiffiffiffiffi . > > < 1 n; α ðvÞ ¼ rffiffiffiffiffiffiffiffiffi . > > : 2 n

u¼0 ; otherwise

ð4Þ

v¼0 ; otherwise

Sub-image is reconstructed from the transformed sub-image block Fb (u,v) by applying f b ðx; yÞ ¼ αðuÞαðvÞ

m−1 n−1 X X u¼0 v¼0

 F b ðu; vÞcos

   ð2x þ 1Þuπ ð2y þ 1Þvπ cos 2m 2n

ð5Þ

for x=0,1,2,…..,m-1 and y=0,1,2,…..,n-1 and ∝ is defined as in Eq. 4. Block based DCT produces three different frequency bands namely low, middle and high frequency bands. In general modification of low frequency band distorted the perceptual quality of the image as it contains maximum image information while the high frequency band can be removed for compression purpose. That is why DCT based watermarking scheme is actually developed on using middle band frequency as it is less perceptible on modification. The top upper left DCT component of block based DCT image is F (0, 0) is the average

Multimed Tools Appl

intensity of the image and also known as the DC coefficient/energy of the image. The other components of the DCT image are called AC coefficients with different low level values. As we move from DC coefficients to AC coefficients deeper in zigzag order, the significant information of the image decreases. Therefore few AC coefficients including DC coefficients are sufficient to represent an approximate image. Figure 3 shows which coefficients pairs are selected for inserting repetition code in our proposed work.

2.4 The repetition code In coding theory, one of the frequently used basic error-correcting codes is repetition code. The idea behind the repetition code is very simple; it only repeats every message bit several times. For its repetition, the convention is that while transmitting messages through a noisy channel it only corrupts minority of these repetitions. So the receiver can recover the message by the principle of majority vote [17] where it basically decodes the message by the corresponding bit value which occurs most often. In this case, the decoded message bit zero (one) is considered when a bit stream consists of more zeros (ones) than ones (zeros). The different kind of repetition codes are shown in Table 1.

3 Proposed scheme In this section, RDWT-DCT-based proposed watermarking scheme is elaborated in details.

Fig. 3 Selected DCT coefficients pair of m × n image-block according to zigzag scanning order

Multimed Tools Appl

Table 1 The Repetition code Repetition code

Information Bit

Code word

(3,1)

0

000

(5,1)

1 0

111 00000

(7,1)

1 0

11111 0000000

1

1111111

3.1 Watermark embedding process The watermark embedding process is depicted in Fig. 4a. Algorithm 1: Watermark Embedding Input: A cover image and a binary logo Output: The watermarked image Begin

(a)

(b)

Fig. 4 a Watermark embedding process. b Watermark bit embedding process

Multimed Tools Appl

Preprocessing for watermark embedding: In this proposed blind image watermarking method to embed binary watermark into grayscale image, preprocessing steps are as follows: Step 1: Decompose the grayscale cover image of size M × N into nonoverlapping blocks of size m × n. Step 2: Apply block level two dimensional RDWT in each nonoverlapping blocks and select LH subband for watermark embedding purpose. Step 3: Apply block level two dimensional DCT in each nonoverlapping LH subband block and select 2 t number of middle band coefficients according to the zigzag scanning order (as shown in Fig. 1) for the purpose of embedding t-bit repetition code. These coefficient pair values are preserved as a secret key. Step 4: For extra security, binary watermark logo is scrambled using Arnold chaotic map. Watermark embedding: Proposed watermark embedding technique is described in below step: Step 5: For each bit of scrambled binary logo, embed the corresponding t-bit repetition code according to the following rules (as shown in Fig. 4b): Rule 1: If the inserting binary logo bit= 1, using t-bit repetition code, make the logo bit t times 1. Then select t-pairs of DCT coefficients from middle band. In each pair, check if 1st coefficient is less than equal to 2nd coefficient, and then swap both these values. Otherwise keep the coefficients unchanged. Rule 2: If the inserting binary logo bit=0, using t-bit repetition code, make the logo bit t times 0. Then select t-pair of DCT coefficients from middle band. In each pair, check if 1st coefficient is greater than 2nd coefficient, and then swap both these values. Otherwise keep the coefficients unchanged. Post processing after watermark embedding: After embedding scrambled binary logo, to get watermarked image post-processing steps are as follows: Step 6: Perform the inverse RDWT on block level after applying inverse DCT on each altered LH subband block. Step 7: Merging all the modified blocks into one block to reconstruct the watermarked image. END

3.2 Watermark extraction process The watermark extraction process is shown in Fig. 5a. Algorithm 1: Watermark Extracting Input: A modified/attack image Output: Watermark logo Begin

Multimed Tools Appl

(a)

(b)

Fig. 5 a Watermark extracting process. b Watermark bit extracting process

Preprocessing for watermark extracting: In this proposed blind image watermarking method to extract the binary watermark from modified/attack image, preprocessing steps are as follows: Step 1: Decompose the grayscale modified/attack image of size M × N into non-overlapping blocks of size m × n. Step 2: Apply block level two dimensional RDWT in each nonoverlapping block and select LH subband for watermark extracting purpose. Step 3: Apply block level two dimensional DCT in each nonoverlapping LH subband block and select 2 t number of middle band coefficients according to the zigzag scanning order. Here coefficient pair values are selected as preserved in the embedding side. Actually, these coefficient pair values are taken from secret key values. Secret key values are transmitted with the watermarked image as extra payload. Watermark extracting: Proposed watermark extracting technique is described in below steps: Step 4: From each block, find out the t-bit codeword from t-pair of selected DCT coefficients according to the following rules (as shown in Fig. 5b): Rule 1: If the intensity of 1st coefficient of the DCT pair is greater than equal to the intensity of 2nd coefficient then the codeword bit= 1. Rule 2: If the intensity of 1st coefficient of the DCT pair is less than the intensity of 2nd coefficient then the codeword bit=0. Step 5: Compute the majority vote for each codeword to establish whether the original watermark bit is a 0 or a 1.

Multimed Tools Appl

Step 6: Obtain the scrambled watermark logo after reshaping the extracted watermark bit streams into a 2-D matrix form. To restore the original watermark, apply the inverse Arnold scrambling transformation to the scrambled logo. END

4 Experimental result To evaluate the performance of the proposed schemes a number of experiments are performed in the MATLAB platform on four standard grayscale benchmark images of size 512 × 512, namely Lena, Airplane, Barbara and Pepper images, one 64 × 64 binary logo and one 128 × 32 binary logo taken as watermarks (as shown in Fig. 6). In this blind image watermarking technique, 8 × 8 block based RDWT is employed on each grayscale image. In the following step, DCT is applied to LH subband of host image and subsequently from each transformed block, according to the zigzag scanning order, middle band AC coefficients from 19thposition onwards up to 2 t position (where t denotes the length of the repetition code) are selected to insert scrambled watermark. The scrambled watermark (as shown in Fig. 6) is embedded to provide extra security. To show the effectiveness of the proposed scheme, proposed RDWT-DCT based watermarking scheme is verified against various experiments in terms of (i) Imperceptibility/Invisibility,(ii) Robustness,(iii) embedding capacity and typical value of t is taken as 5, 7, 9 and 11 respectively.

Imperceptibility measurement To calculate the imperceptibility/invisibility measurement, alteration of perceptual image quality (by the proposed watermarking method) should be determined. The peak signal-to-noise-ratio (PSNR) is utilized to find perceptual similarity between a host image and a watermarked image. In an effective invisible watermarking algorithm (i) watermark should be imperceptible/invisible from HVS and should check with (ii) standard benchmark PSNR. PSNR values are presented in decibel (dB). For optimized imperceptibility, the minimum acceptable value of PSNR is 38 dB as suggested by petitcolas [19]. This convention is dubious because PSNR is not a meaningful constraint in the context of geometric distortions [51]. PSNR can be defined as follows: PSNR ¼ 10  log10

maxðxði; jÞÞ2 MSE

ð6Þ

Where, the mean square error (MSE) between host image x and the watermarked image x is defined as follows: 2 XM XN  1 xi j − xi j ð7Þ MSE ¼ i¼1 j¼1 M N

Fig. 6 a–d Original Test Images e The binary watermark of size 64 × 64 f The scrambled watermark g The binary watermark of size 128 × 32 h The scrambled watermark

Multimed Tools Appl

Here, the notations M and N represent the width and height of an image, xij Is the pixel intensity value of coordinate (i, j) of the original image and xi j is corresponding value of watermarked image. Basically, PSNR has been computed to compare the visual quality between cover image and watermarked image after embedding the watermark. Table 2 summarizes the experimental results for the proposed watermarking scheme in context of MSE and PSNR value without any modification/attack of watermarked image.

Robustness measurement The robustness indicates that the watermarked object should resist against some watermark removal intentional/unintentional attack. To measure the robustness property of proposed method, bit error rate (BER) and normalized cross-correlation (NC) value between the original watermark and extracted distorted watermark (without attack/after applying different types of attack) is compared. The range of NC value lies between −1 to +1. If watermarked image is almost similar like the original image, then this correlation value is approximately 1 while −1 correlation value indicates negative like watermarked image. It becomes totally unacceptable or uncorrelated, if the NC value tends to 0. The BER and NC can be calculated as follows: BER ¼

Number of error bits Number of error bits per second ¼ Total bits transmitted Data rate per second

ð8Þ

  ½ w ð i; j Þ − μ   w ð i; j Þ − μ  w i¼1 j¼1 w NC w; w ¼ rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi  2 X X XM XN M N 2 ½wði; jÞ − μw   wði; jÞ−μ i¼1 j¼1 i¼1 j¼1 XM XN

ð9Þ

w

Where M and N represent the width and height of the watermark image, w(i, j) = the pixel intensity value at coordinate (i, j) of original watermark, wði; jÞ =the pixel intensity value at coordinates (i, j) of extracting watermark, μw = mean of the original watermark, μw =mean of the extracted watermark respectively. BER and NC value of extracting watermark is calculated without any modification/attack of watermarked image. These values are presented in Table 3. All the watermarked images using various repetition codes and recovered watermark logos from the corresponding watermarked images are shown side by side in Fig. 7. As a representative, throughout this paper only experimental result of Pepper images are presented with the 64 × 64 binary logo. Table 2 MSE and PSNR of watermarked images using 5 Repetition Code,7 Repetition Code,9 Repetition Code and 11 Repetition Code without attack

Image name

Lena Airplane Barbara Pepper

Using 5 repetition code

Using 7 repetition code

Using 9 repetition code

Using 11 repetition code

MSE

PSNR

MSE

PSNR

MSE

PSNR

MSE

PSNR

0.4648 1.2189 1.0461 1.3162

51.4581 47.3052 47.9349 46.9716

0.4902 1.2450 1.3671 1.3181

51.2608 47.2132 46.8068 46.9652

0.5117 1.2966 1.5379 1.3333

51.0743 47.0369 46.2955 46.9157

0.9273 2.0986 2.1803 2.5704

48.4928 44.9115 44.7797 44.0648

Multimed Tools Appl Table 3 NC and BER between the original watermark image and extracted watermark image from cover image using 5 Repetition Code,7 Repetition Code,9 Repetition Code and 11 Repetition Code without attack

Image name

Using 5 repetition code

Using 7 repetition code

Using 9 repetition code

Using 11 repetition code

NC

BER

NC

BER

NC

BER

NC

BER

Lena

0.9901

0.0088

0.9991

0.0015

0.9994

0.0005

1

0.0004

Airplane Barbara Pepper

0.9883 0.9955 0.9982

0.0125 0.0039 0.0029

0.9910 0.9986 0.9997

0.0085 0.0019 0.0001

0.9946 0.9993 1

0.0046 0.0007 0

1 1 1

0.0003 0.0001 0

Embedding capacity measurement Amount of data embedded as watermark to extract the watermark effectively in receiver side without affecting imperceptibility of original data is termed as the capacity of watermark or watermark payload. By increasing the embedded watermarking data capacity enhanced the watermarking scheme robustness which may affect the imperceptibility of the watermarked data. In this proposed scheme, a binary bit of watermark is embedded into each 8 × 8 decomposed and RDWT-DCT transformed host image non-overlapping blocks by modifying some middle significant AC coefficient pairs using repetition code. By increasing the repetition code enhanced the watermarking scheme robustness (as NC and BER values are shown in Table 3) which may affect the imperceptibility (as MSE and PSNR values are shown in Table 2) of the watermarked image. So there need a negotiation between these three properties to design an efficient watermarking scheme. The proposed scheme’s embedding capacity can be adjusted by varying the host image’s decomposition block size. Maximum watermark payload of this technique is presented in Table 4 with various parameters of the cover image. During transmission over the Internet, the watermarked image may be manipulated in an illicit manner. For performance evaluation and fair benchmarking, the proposed technique is verified against various attacks. In [51], Stirmark benchmark is represented as a watermarking scheme benchmark where different types of image watermarking scheme attacks are divided into some categories like signal enhancement, scaling, cropping, shearing, rotation, linear transformations, other geometric transformations, compression and combined attacks. Kutter and Petitcolas [19] divide these attacks into some sections according to their properties. The proposed scheme is tested against these attacks with all the test images and 64 × 64 logo. In addition to this, robustness performance of proposed scheme is evaluated with standard benchmark software BCheckmark^ [1, 35] which is shown in Table 12. Checkmark attacks are applied on Lena test image of size 512 × 512 and binary logo of size 128 × 32.

Fig. 7 Watermarked Pepper image and recovered watermark without attack using a 5 Repetition code b 7 Repetition code c 9 Repetition code d 11 Repetition code

Multimed Tools Appl

Table 4 Maximum watermark payload of proposed scheme with various parameters of cover image

Size of Cover Image

Decomposed Block Size of Host Image

Maximum Payload/Watermark Size in Bits

512 × 512

8×8

4096/212

512 × 512 256 × 256 256 × 256

4×4 8×8 4×4

16384/214 1024/210 4096/212

4.1 Enhancement technique attacks (i) Low pass filtering: Lowpass filtering operation is used to remove high-frequency noise from a signal. To prove the robustness of the proposed technique against low pass filter, gaussian lowpass filtering attack with two different size parameters (2 × 2,3 × 3) are applied on selected standard test images. Modified watermarked image under Gaussian lowpass filtering attack (2 × 2) and its corresponding recovered watermark logos are presented in Fig. 14a. Recovered logos using 5 repetition code are first row first column logos. Recovered logos using 7 repetition code are first row second column logos. Recovered logos using 9 repetition code are second row first column logos. Recovered logos using 11 repetition code are second row second column logos and this presentation rule is followed for other recovered logos also (as presented in Fig. 14). BER and NC values of recovering watermark logos from this attack are tabulated in Table 5 for four different test images. All the rows for Gaussian lowpass filtering attack in this table conclude that recovering watermark data of this proposed method survive the gaussian lowpass manipulation filtering attack and PSNR comparison diagram in Fig. 8 prove the imperceptibility property of this method. (ii) Median filtering: In common image enhancement application, a median filter is not often used to achieve blurring rather than it is used to the noise reduction process. Basically, median filter modifies the center pixel value of the window with the middle value of the sorted pixel values. The proposed scheme is examined against median filtering attacks with different window size (2 × 2,3 × 3). Recovered watermark logos using 5 repetition code,7 repetition code,9 repetition code,11 repetition code and its corresponding median filtering (2 × 2) modified image are shown in Fig. 14b. BER and NC values of recovering watermark for all the test images (as presented in Table 5), prove the robustness and PSNR plot in Fig. 8 confirm the steadiness of this scheme against median filtering attack. (iii) Wiener filtering: To simulate the effects of two-dimensional linear predictive image coding, wiener filter is applied. With the improvement of prediction filter order size, the model becomes closer to the watermarked image block. So, estimation based attack become easier, if the attacker has some basic knowledge about the watermark’s embedding process. Image denoising is one of the malicious estimation based attack. This technique is experimented against wiener filtering attack under different size parameters. Wiener filtering modified image and corresponding logos using 5 repetition code,7 repetition code,9 repetition code,11 repetition code are depicted in Fig. 14c. Analysis of wiener filter rows in Table 5 exhibits the consistency of the BER and NC values for recovering watermark data where PSNR graph in Fig. 8 shows the imperceptibility against wiener filtering. (iv) Average filtering: Average filtering is one of the well known image enhancement technique attack as well as one of the denoising attacks. An average filter is applied to

Multimed Tools Appl Table 5 NC and BER comparison of extracted watermark using 5 Repetition Code,7 Repetition Code,9 Repetition Code and 11 Repetition Code under various attacks

Attack

Using 5 repetition code

Using 7 repetition code

Using 9 repetition code

Using 11 repetition code

NC

NC

NC

NC

BER

BER

BER

BER

Lena Gaussian Lowpass Filter(2,2) 0.8779 0.1450 0.8818 0.1427 0.9349 0.0679 0.9555 0.0457 Gaussian Lowpass Filter(3,3) 0.9878 0.0146 0.9946 0.0059 0.9982 0.0022 0.9991 0.0004 Median Filter(2,2) 0.8512 0.1641 0.8597 0.1602 0.9124 0.0928 0.9404 0.0610 Median Filter(3,3) Wiener Filter(2,2) Wiener Filter(3,3)

0.7366 0.2874 0.7374 0.2870 0.7400 0.2853 0.7625 0.2715 0.9005 0.1140 0.9143 0.0969 0.9547 0.0461 0.9682 0.0310 0.7433 0.2804 0.7500 0.2786 0.7556 0.2769 0.7864 0.2429

Average Filter(2,2) Average Filter(3,3)

0.8779 0.1450 0.8718 0.1487 0.9349 0.0679 0.9555 0.0457 0.7162 0.3171 0.7268 0.3096 0.7321 0.3064 0.7226 0.3142

Airplane Gaussian Lowpass Filter(2,2) 0.8629 0.1597 0.8645 0.1590 0.9292 0.0789 0.9563 0.0466 Gaussian Lowpass Filter(3,3) 0.9832 0.0225 0.9879 0.0112 0.9941 0.0054 0.9959 0.0037 Median Filter(2,2)

0.8451 0.1741 0.8452 0.1770 0.9081 0.1006 0.9328 0.0691

Median Filter(3,3) Wiener Filter(2,2)

0.7533 0.2766 0.7562 0.2756 0.7573 0.2754 0.7774 0.2493 0.9010 0.1157 0.9084 0.1099 0.9497 0.0608 0.9691 0.0366

Wiener Filter(3,3) Average Filter(2,2)

0.7363 0.2976 0.7389 0.2932 0.7575 0.2759 0.7789 0.2463 0.8629 0.1597 0.8636 0.1594 0.9293 0.0786 0.9563 0.0466

Average Filter(3,3) Barbara Gaussian Lowpass Filter(2,2) Gaussian Lowpass Filter(3,3) Median Filter(2,2) Median Filter(3,3) Wiener Filter(2,2) Wiener Filter(3,3) Average Filter(2,2) Average Filter(3,3) Pepper Gaussian Lowpass Filter(2,2) Gaussian Lowpass Filter(3,3) Median Filter(2,2) Median Filter(3,3) Wiener Filter(2,2) Wiener Filter(3,3) Average Filter(2,2) Average Filter(3,3)

0.7201 0.3165 0.7348 0.2996 0.7418 0.2894 0.7662 0.2632 0.8695 0.1536 0.8730 0.1507 0.9356 0.0754 0.9606 0.0498 0.9896 0.0115 0.9977 0.0027 0.9995 0.0004 0.9995 0.0004 0.8330 0.1951 0.8339 0.1904 0.8910 0.1238 0.9267 0.0891 0.7253 0.8968 0.7604 0.8695

0.3030 0.1182 0.2607 0.1536

0.7323 0.9086 0.7648 0.8730

0.2993 0.1042 0.2620 0.1519

0.7542 0.9557 0.7893 0.9356

0.2766 0.0540 0.2390 0.0754

0.7729 0.9706 0.8186 0.9606

0.2527 0.0327 0.2039 0.0498

0.7303 0.3005 0.7376 0.2994 0.7502 0.2767 0.7557 0.2744 0.8898 0.1296 0.8948 0.1277 0.9465 0.0681 0.9736 0.0322 0.9964 0.0066 0.9986 0.0022 0.9995 0.0004 1 0.0004 0.8596 0.1619 0.8602 0.1597 0.9271 0.0891 0.9509 0.0605 0.7138 0.9144 0.7344 0.8898 0.7178

0.3432 0.1008 0.2880 0.1296 0.3417

0.7286 0.9159 0.7464 0.8948 0.7256

0.3113 0.1003 0.2803 0.1277 0.3128

0.7338 0.9642 0.7551 0.9465 0.7372

0.3230 0.0464 0.2717 0.0681 0.3108

0.7452 0.9832 0.7890 0.9736 0.7398

0.2876 0.0227 0.2297 0.0322 0.2959

Multimed Tools Appl PSNR comparison of pepper image under various aack

60 40 20 0 Gaussian Lowpass Gaussian Lowpass Median Filter(2,2) Median Filter(3,3) Wiener Filter (2,2) Wiener Filter (3,3) Average Filter (2,2) Average Filter (3,3) Filter(2,2) Filter(3,3) USING 5 REPETITION CODE

USING 7 REPETITION CODE

USING 9 REPETITION CODE

USING 11 REPETITION CODE

Fig. 8 Imperceptibilty verification of proposed scheme under various attacks

replace each sample of the watermarked image with the average value from the set of W × W neighboring pixels or W window size. This scheme is verified against average filtering attack and one average filtering attack image and its corresponding watermarks using 5 repetition code,7 repetition code,9 repetition code,11 repetition code are shown in Fig. 14d. NC and BER values of recovering watermark data demonstrate the inverse proportionality and proportionality with the window size but for repetition code values it are opposite. This is provided as a tabular format in Table 5. For invisibility verification, the PSNR comparison diagram is presented in Fig. 8 under this attack. (v) Image sharpening: Sharpening attack is used to detect high frequency noise introduced by unauthorized users. This proposed method is tested against sharpening operation as shown in Fig. 14e. NC and BER values of recovering watermark data for all the test images tabulate in Table 6 where NC is proportional to repetition code value and BER is inverse proportional. Imperceptibility evaluation is presented in Fig. 9 with PSNR values under sharpening attack. (vi) Histogram modification: To compensate clarity and brightness of an image, sometimes histogram of the image is modified. The histogram modification enhancement technique is common to improve the visual quality of any image. The proposed technique is tested against histogram modification operation as shown in Fig. 14f. Histogram equalization rows in Table 6 demonstrate the proportionality of NC and inverse proportionality BER values of recovering watermark for all test images. Imperceptibility property of proposed scheme can be validated easily by the PSNR comparison chart in Fig. 9. (vii) Gamma correction: Gamma correction is one of the popular image enhancement techniques to adjust poor image display quality. Also, sometimes intentionally or unintentionally image is enhanced by power law transformation or gamma correction method. Figure 14g and h show watermarked image and its corresponding watermarks using 5 repetition code,7 repetition code,9 repetition code,11 repetition code after applying gamma correction with two different gamma values. NC and BER values of recovering watermark data put in Table 6. Gamma correction rows for all the test images in the above mentioned table and PSNR comparison illustration in Fig. 9 provide enough evidence to establish the stability of the proposed method against this attack.

4.2 Noise addition attack (viii) Noise addition: Most addressed non-geometrical attack in signal processing is the addition of additive noise and uncorrelated multiplicative noise. The proposed scheme is experimented against salt & peppers, speckle and Gaussian noise. Imperceptibility property of this scheme is protected against noise attacks. It is presented in Fig. 10.

0.9874 0.9837

Gamma attack(gamma=0.5) Gamma attack(gamma=0.25)

0.9829 0.9757 0.9621

Gamma attack(gamma=0.75) Gamma attack(gamma=0.5) Gamma attack(gamma=0.25)

0.9855 0.9656

0.9909

0.9937

Histogram Equalization Gamma attack(gamma=3)

Gamma attack(gamma=2)

Gamma attack(gamma=0.75)

0.9850

0.9833

Gamma attack(gamma=2)

Barbara Image Sharpening

0.9782

Gamma attack(gamma=3)

0.9815 0.9588

0.9336 0.9832 0.9892

Gamma attack(gamma=3) Gamma attack(gamma=2) Gamma attack(gamma=0.75)

Airplane Image Sharpening Histogram Equalization

0.9837 0.9729

0.0063

0.0139

0.0186 0.0444

0.0212

0.0183 0.0249 0.0405

0.0173

0.0273

0.0198 0.0454

0.0127 0.0186

0.0945 0.0212 0.0100

0.0176 0.0295

0.9995

0.9968

0.9951 0.9837

0.9955

0.9901 0.9896 0.9796

0.9905

0.9873

0.9900 0.9764

0.9959 0.9928

0.9568 0.9928 0.9968

0.9955 0.9874

0.0004

0.0044

0.0061 0.0208

0.0054

0.0098 0.0117 0.0242

0.0093

0.0159

0.0117 0.0288

0.0042 0.0085

0.0500 0.0090 0.0029

0.0051 0.0129

BER

NC

NC

BER

Using 7 repetition code

Using 5 repetition code

Image Sharpening Histogram Equalization

Lena

Attack

0.9992

0.9982

0.9957 0.9863

0.9950

0.9937 0.9919 0.9842

0.9941

0.9918

0.9946 0.9854

0.9991 0.9937

0.9656 0.9928 0.9995

0.9977 0.9946

NC

0.0004

0.0027

0.0051 0.0198

0.0049

0.0054 0.0073 0.0183

0.0068

0.0117

0.0066 0.0212

0.0012 0.0056

0.0444 0.0066 0.0004

0.0022 0.0068

BER

Using 9 repetition code

1

0.9995

0.9991 0.9918

0.9977

0.9964 0.9959 0.9887

0.9977

0.9941

0.9973 0.9891

0.9995 0.9950

0.9700 0.9932 1

0.9995 0.9950

NC

0.0001

0.0004

0.0019 0.0127

0.0020

0.0029 0.0042 0.0117

0.0029

0.0063

0.0029 0.0129

0.0004 0.0046

0.0364 0.0059 0

0.0004 0.0061

BER

Using 11 repetition code

Table 6 NC and BER comparison of extracted watermark using 5 Repetition Code,7 Repetition Code,9 Repetition Code and 11 Repetition Code under various attacks

Multimed Tools Appl

0.9635 0.9840 0.9968

0.9964

0.9896

Gamma attack(gamma=3) Gamma attack(gamma=2) Gamma attack(gamma=0.75)

Gamma attack(gamma=0.5)

Gamma attack(gamma=0.25)

0.0088

0.0117

0.0046

0.0454 0.0237 0.0044

0.0225

0.9928

0.9801

0.0137

0.9860

Histogram Equalization

0.0061

0.9928

Gamma attack(gamma=0.5)

Gamma attack(gamma=0.25) Pepper Image Sharpening

0.9955

0.9986

0.9772 0.9913 0.9995

0.9941

0.9982

0.9977

0.9991

NC

BER

NC

0.0042

0.0015

0.0295 0.0146 0.0004

0.0095

0.0017

0.0032

0.0015

BER

Using 7 repetition code

Using 5 repetition code

Attack

Table 6 (continued)

0.9955

0.9991

0.9853 0.9909 1

0.9950

0.9995

0.9979

0.9991

NC

0.0044

0.0004

0.0237 0.0156 0

0.0056

0.0004

0.0027

0.0004

BER

Using 9 repetition code

0.9973

0.9991

0.9845 0.9941 1

0.9964

0.9995

0.9982

1

NC

0.0039

0.0004

0.0217 0.0117 0

0.0039

0.0004

0.0017

0.0001

BER

Using 11 repetition code

Multimed Tools Appl

Multimed Tools Appl PSNR comparison of pepper image under various aack

40 20 0 Image Sharpening

Histogram Equalisation USING 5 REPETITION CODE

Gamma=3 attack

Gamma=2 attack

USING 7 REPETITION CODE

Gamma=0.75 attack

USING 9 REPETITION CODE

Gamma=0.5 attack

Gamma=0.25 attack

USING 11 REPETITION CODE

Fig. 9 Imperceptibility verification of proposed scheme under various attacks

Salt and pepper noise is caused by pixel’s error at the time of data transmission. In salt and pepper noise, corrupted pixel’s values are either set to zero or maximum value or single bits flipped over. This pixel’s value modification gives the image salt and pepper like appearance where noise density is calculated by alteration of percentage of pixels. This method is tested against salt and pepper noise with four different noise density as presented in the Table 7. Salt & pepper noise modified watermarked image with noise density 0.01 is shown in Fig. 14i. In salt and pepper noise, inverse proportionality of NC values and proportionality of BER values of recover watermark with noise density are given in Table 7 where NC and BER values show reverse proportionality property in case of repetition code. Speckle noise is one of the multiplicative noises where speckle exists inherently as a granular noise. The variance of a single pixel is equal to the variance of the local area that is centered on that pixel. This method is tested against speckle noise under four different noise variances. From Table 7, it can be easily understood that recovering watermark logo quality is improved with the decrement of noise variance. NC values of recovering watermark exhibit inverse proportionality with speckle image noise variance where BER shows proportionality characteristic, but this proportionality property is turned around in case of repetition code. Manipulated watermarked image by speckle noise with variance=0.01 is portrayed in Fig. 14j. Gaussian noise is one of the commonly used statistical noise processing operations. The amount of noise is varied by its variance with zero mean. For robustness clarification, this statistical noise is added to the watermarked images of this method. Addition of gaussian noise under different variances (with mean=0) are depicted in Table 7 for all the test images. These NC PSNR comparison of pepper image under various aack

40 35 30 25 20 15 10 5 0

USING 5 REPETITION CODE

USING 7 REPETITION CODE

USING 9 REPETITION CODE

USING 11 REPETITION CODE

Fig. 10 Imperceptibility verification of proposed scheme under noise addition attacks

0.8446 0.9084

0.9489 0.9746 0.7402

0.7878 0.8384

0.8978 0.7369 0.8093

0.8282

0.9011

0.8417 0.9186

0.9351 0.9663

0.7085 0.7452

0.7611

0.8082

Salt & Pepper noise(density=0.003) Salt & Pepper noise(density=0.001) Speckle noise (var=0.01)

Speckle noise (var=0.005) Speckle noise (var=0.003)

Speckle noise (var=0.001) Gaussian noise(M=0,var=0.01) Gaussian noise(M=0, var=0.005)

Gaussian noise(M=0, var=0.003)

Gaussian noise(M=0, var=0.001)

Airplane Salt & Pepper noise(density=0.01) Salt & Pepper noise(density=0.005)

Salt & Pepper noise(density=0.003) Salt & Pepper noise(density=0.001)

Speckle noise (var=0.01) Speckle noise (var=0.005)

Speckle noise (var=0.003)

Speckle noise (var=0.001)

0.2104

0.2712

0.3291 0.2856

0.0664 0.0334

0.1665 0.0911

0.1130

0.2041

0.1133 0.2932 0.2192

0.2368 0.1890

0.0649 0.0305 0.2888

0.1738 0.1033

0.8269

0.7840

0.7168 0.7522

0.9509 0.9764

0.8690 0.9249

0.9263

0.8649

0.9198 0.7634 0.8236

0.8333 0.8605

0.9564 0.9809 0.8038

0.8627 0.9198

0.1904

0.2505

0.3154 0.2783

0.0564 0.0273

0.1528 0.0842

0.0918

0.1545

0.0962 0.2668 0.2041

0.1931 0.1624

0.0554 0.0247 0.2324

0.1616 0.0933

BER

NC

NC

BER

Using 7 repetition code

Using 5 repetition code

Salt & Pepper noise(density=0.01) Salt & Pepper noise(density=0.005)

Lena

Attack

0.8423

0.7917

0.7190 0.7650

0.9437 0.9814

0.8641 0.9250

0.9402

0.8794

0.9283 0.7834 0.8439

0.8437 0.8738

0.9538 0.9846 0.8079

0.8616 0.9259

NC

0.1760

0.2351

0.3147 0.2617

0.0554 0.0215

0.1470 0.0825

0.0630

0.1445

0.0796 0.2527 0.1880

0.1724 0.1472

0.0559 0.0181 0.2183

0.1631 0.0886

BER

Using 9 repetition code

0.8569

0.7961

0.7382 0.7661

0.9560 0.9859

0.8863 0.9418

0.9499

0.8902

0.9415 0.8152 0.8531

0.8563 0.8894

0.9576 0.9823 0.8309

0.8745 0.9314

NC

0.1658

0.2324

0.2915 0.2637

0.0503 0.0188

0.1292 0.0667

0.0537

0.1277

0.0657 0.2161 0.1616

0.1631 0.1282

0.0508 0.0220 0.1997

0.1501 0.0833

BER

Using 11 repetition code

Table 7 NC and BER comparison of extracted watermark using 5 Repetition Code,7 Repetition Code,9 Repetition Code and 11 Repetition Code under various attacks

Multimed Tools Appl

0.1443 0.0837 0.1655 0.0820

0.8580 0.9135 0.8253 0.8503

0.8784 0.9301

0.8623 0.9339

0.9615 0.9823

Speckle noise (var=0.003) Speckle noise (var=0.001) Gaussian noise(M=0,var=0.01) Gaussian noise(M=0, var=0.005)

Gaussian noise(M=0, var=0.003) Gaussian noise(M=0, var=0.001) Pepper Salt & Pepper noise(density=0.01) Salt & Pepper noise(density=0.005)

Salt & Pepper noise(density=0.003) Salt & Pepper noise(density=0.001)

0.0496 0.0217

0.1616 0.0999 0.1980 0.1672

0.2236 0.1868

0.0234

0.9728

0.8017 0.8401

Speckle noise (var=0.01) Speckle noise (var=0.005)

0.1548 0.0828 0.0493

0.1995 0.1794 0.1177

0.2322

Salt & Pepper noise(density=0.001)

0.8566 0.9264 0.9528

Salt & Pepper noise(density=0.01) Salt & Pepper noise(density=0.005) Salt & Pepper noise(density=0.003)

Barbara

0.7973

0.8248 0.8458 0.8988

Gaussian noise(M=0, var=0.005) Gaussian noise(M=0, var=0.003) Gaussian noise(M=0, var=0.001)

0.9632 0.9833

0.8752 0.9293

0.9158 0.9612

0.8900 0.9356 0.8541 0.8933

0.8209 0.8703

0.9882

0.8882 0.9451 0.9628

0.8267 0.8582 0.9223

0.7902

0.0498 0.0186

0.1487 0.0813

0.0969 0.0474

0.1255 0.0720 0.1650 0.1233

0.2004 0.1489

0.0144

0.1306 0.0698 0.0481

0.1975 0.1560 0.0894

0.2349

BER

NC

NC

BER

Using 7 repetition code

Using 5 repetition code

Gaussian noise(M=0,var=0.01)

Attack

Table 7 (continued)

0.9671 0.9797

0.8744 0.9398

0.9233 0.9702

0.8854 0.9354 0.8441 0.8930

0.8312 0.8689

0.9901

0.8882 0.9499 0.9639

0.8380 0.8768 0.9343

0.7992

NC

0.0417 0.0205

0.1418 0.0698

0.0938 0.0405

0.1287 0.0698 0.1702 0.1160

0.1863 0.1506

0.0112

0.1289 0.0718 0.0454

0.1821 0.1418 0.0791

0.2244

BER

Using 9 repetition code

0.9725 0.9928

0.8941 0.9496

0.9245 0.9706

0.9051 0.9443 0.8641 0.8974

0.8326 0.8719

0.9923

0.9080 0.9505 0.9687

0.8642 0.8899 0.9449

0.8197

NC

0.0374 0.0093

0.1284 0.0620

0.0850 0.0334

0.1052 0.0618 0.1521 0.1152

0.1848 0.1411

0.0107

0.1099 0.0625 0.0430

0.1543 0.1223 0.0679

0.2109

BER

Using 11 repetition code

Multimed Tools Appl

0.8111

0.8549 0.8944 0.9564

Gaussian noise(M=0,var=0.01)

Gaussian noise(M=0, var=0.005) Gaussian noise(M=0, var=0.003) Gaussian noise(M=0, var=0.001)

0.1716 0.1216 0.0554

0.2158

0.1802 0.1523 0.0842 0.8831 0.9149 0.9691

0.8343

0.8638 0.8906 0.9448

0.8089

0.2358

0.7964

0.8512 0.8698 0.9301

Speckle noise (var=0.01)

NC

BER

NC

0.1326 0.0979 0.0364

0.1960

0.1572 0.1348 0.0649

0.2141

BER

Using 7 repetition code

Using 5 repetition code

Speckle noise (var=0.005) Speckle noise (var=0.003) Speckle noise (var=0.001)

Attack

Table 7 (continued)

0.9030 0.9356 0.9826

0.8589

0.8791 0.9121 0.9578

0.8371

NC

0.1106 0.0781 0.0271

0.1682

0.1414 0.1150 0.0530

0.1851

BER

Using 9 repetition code

0.9248 0.9512 0.9838

0.8631

0.8886 0.9167 0.9543

0.8421

NC

0.0938 0.0623 0.0173

0.1587

0.1284 0.0959 0.0503

0.1812

BER

Using 11 repetition code

Multimed Tools Appl

Multimed Tools Appl Table 8 NC and BER comparison of extracted watermark using 5 Repetition Code,7 Repetition Code,9 Repetition Code and 11 Repetition Code under various attacks

Attack

Using 5 repetition code

Using 7 repetition code

Using 9 repetition code

Using 11 repetition code

NC

BER

NC

BER

NC

BER

NC

BER

0.8414 0.8512 0.9391 0.9701 0.9887

0.1790 0.1672 0.1060 0.0190 0.0125

0.8705 0.8792 0.8987 0.9991 0.9968

0.1524 0.1406 0.1199 0.0020 0.0054

0.8883 0.8936 0.9856 1 0.9991

0.1324 0.1195 0.0154 0.0004 0.0044

0.8941 0.9152 0.9869 1 0.9986

0.1182 0.0984 0.0132 0.0004 0.0029

Lena Rotation(clockwise 5°) Rotation(anticlockwise 5°) Cropping(256 × 256 by black) Cropping(256 × 256 by white) Cropping(128 × 128 by black)

Cropping(128 × 128 by white) 0.9843 0.0122 0.9971 0.0032 0.9946 0.0039 0.9955 0.0029 Cropping(64 × 64 by black) Cropping(64 × 64 by white) Airplane Rotation(clockwise 5°) Rotation(anticlockwise 5°)

0.9897 0.0093 0.9986 0.0017 0.9995 0.0004 1 0.9901 0.0098 0.9901 0.0022 0.9995 0.0004 1

0.0001 0.0001

0.8403 0.1786 0.8694 0.1529 0.8872 0.1331 0.8958 0.1190 0.8502 0.1675 0.8781 0.1415 0.8925 0.1203 0.9142 0.0989

Cropping(256 × 256 by black) 0.9685 0.0396 0.9730 0.0271 0.9837 0.0186 0.9860 0.0144 Cropping(256 × 256 by white) 0.9883 0.0127 0.9910 0.0093 0.9946 0.0054 0.9964 0.0039 Cropping(128 × 128 by black) 0.9864 0.0168 0.9896 0.0117 0.9937 0.0088 0.9950 0.0071 Cropping(128 × 128 by white) 0.9811 0.0178 0.9865 0.0129 0.9892 0.0093 0.9932 0.0063 Cropping(64 × 64 by black) 0.9869 0.0132 0.9910 0.0085 0.9946 0.0046 0.9964 0.0032 Cropping(64 × 64 by white) Barbara Rotation(clockwise 5°) Rotation(anticlockwise 5°) Cropping(256 × 256 by black) Cropping(256 × 256 by white) Cropping(128 × 128 by black) Cropping(128 × 128 by white) Cropping(64 × 64 by black) Cropping(64 × 64 by white) Pepper Rotation(clockwise 5°) Rotation(anticlockwise 5°) Cropping(256 × 256 by black) Cropping(256 × 256 by white) Cropping(128 × 128 by black) Cropping(128 × 128 by white) Cropping(64 × 64 by black) Cropping(64 × 64 by white)

0.9883 0.0127 0.9910 0.0085 0.9946 0.0046 0.9964 0.0032 0.8628 0.1618 0.8894 0.1284 0.9057 0.1076 0.9281 0.0874 0.8637 0.1609 0.8901 0.1275 0.9064 0.1071 0.9295 0.0868 0.9745 0.9955 0.9946 0.9901 0.9946

0.0320 0.0039 0.0083 0.0078 0.0051

0.9832 0.9986 0.9964 0.9946 0.9982

0.0208 0.0012 0.0042 0.0042 0.0015

0.9891 0.9991 0.9968 0.9937 0.9991

0.0159 0.0004 0.0061 0.0042 0.0004

0.9896 1 0.9986 0.9977 1

0.9955 0.0044 0.9986 0.0017 0.9991 0.0004 1

0.0115 0.0004 0.0039 0.0022 0.0004 0.0004

0.8617 0.1621 0.8887 0.1289 0.9120 0.0987 0.9312 0.0961 0.8625 0.9882 0.9982 0.9973 0.9915 0.9977 0.9982

0.1618 0.0171 0.0029 0.0051 0.0073 0.0034 0.0042

0.8908 0.9928 1 0.9986 0.9964 0.9995 1

0.1269 0.0078 0 0.0020 0.0024 0.0004 0.0004

0.9102 0.9959 1 0.9982 0.9933 1 1

0.0994 0.0073 0 0.0024 0.0037 0 0

0.9406 0.9969 1 0.9991 0.9964 1 1

0.0674 0.0056 0.0004 0.0015 0.0020 0 0

Multimed Tools Appl

and BER values are consistent for different variances and different repetition codes. Watermarked image with additive gaussian noise (variance=0.01, mean=0) is described in Fig. 14k.

4.3 Geometric transformation attacks (x) Rotation: The commercial value of an image and its small rotated version is not differing too much. But watermark data can be affected by the little bit of rotation. To verify robustness against rotation geometric attack, the proposed technique is tested against rotation attack with 5° angle in clockwise as well as anticlockwise. Figure of anticlockwise rotation attack is presented in Fig. 14l. BER and NC values of recovering watermark are demonstrated in Table 8. Recover watermark images for different repetition codes are not obscure and distinctively represent the watermark logo as seen in Fig. 14l. Imperceptibility compatibility of the proposed scheme against rotation attack is shown in Fig. 11. (xi) Cropping: To capture the required part of an object, cropping/focusing is done on necessitate part and rest part of object or image is neglected. Cropping operation is also used to change aspect ratio of image or to improve image framing. Robustness of the proposed technique is tested against cropping operation by changing the cropping window size. Watermarked images of proposed scheme are experimented against cropping geometric attack under different cropping window sizes. For the proposed method 256 × 256,128 × 128 and 64 × 64 cropping window size is used. This cropping window may be black or white alternatively. Recover watermark image’s NC and BER values provide enough evidence of the stability of the proposed method as demonstrated in Table 8. One cropping attack image where 256 × 256 pixels are replaced by black window is shown in Fig. 14m. Corresponding prominent recover watermark logo using different repetition codes are given in the same Figure. Comparative graph in Fig. 11, verifies the invisibility property of the proposed scheme against cropping attack. (xii) Scaling: Image may be resized intentionally or unintentionally in uniform/non-uniform manner. For example, scaling of the image is a normal scenario for scanning of a hard copy image. So, proposed scheme is tested against scaling operation to prove the robustness and imperceptibilty property against scaling. At first, the image is resized by multiplying by a scaling factor. Then again the image is scaled back to its original size. As a representative, one watermarked image under this scaling geometric attack are represented in Fig. 14n. Recover watermark images under different repetition codes are also given in Fig. 14n. NC PSNR comparison of pepper image under various aack

30 25 20 15 10 5 0

USING 5 REPETITION CODE

USING 7 REPETITION CODE

USING 9 REPETITION CODE

USING 11 REPETITION CODE

Fig. 11 Imperceptibility verification of proposed scheme under various geometric attacks

0.9860 0.9892

0.9910

Scaling(zoomout=2,zoomin=0.5) Scaling(zoomout=4,zoomin=0.25)

Scaling(zoomout=8,zoomin=0.125)

0.1774

0.8411

0.8369 0.8094

Scaling(zoomout=0.25,zoomin==4) Scaling(zoomout=0.125,zoomin=8)

0.0146

0.9869

Scaling(zoomout=8,zoomin=0.125) Barbara Scaling(zoomout=0.5,zoomin=2)

0.0107

0.0149 0.0112

0.1782 0.2191

0.0193

0.0217

0.1599 0.1878 0.2181

0.9819

0.8574 0.8386 0.8126

Scaling(zoomout=0.5,zoomin=2) Scaling(zoomout=0.25,zoomin=4) Scaling(zoomout=0.125,zoomin=8)

0.0142

0.9824

0.9869

Scaling(zoomout=8,zoomin=0.125) Airplane

0.2173 0.0203 0.0129

Scaling(zoomout=4,zoomin=0.25)

0.8142 0.9832 0.9883

Scaling(zoomout=0.125,zoomin=8) Scaling(zoomout=2,zoomin=0.5) Scaling(zoomout=4,zoomin=0.25)

0.1593 0.1872

Scaling(zoomout=2,zoomin=0.5)

0.8583 0.8394

0.9982

0.9964 0.9977

0.8592 0.8392

0.8673

0.9914

0.9905

0.9892

0.8642 0.8497 0.8317

0.9977

0.8339 0.9946 0.9968

0.8650 0.8503

0.0020

0.0042 0.0022

0.1601 0.1884

0.1595

0.0088

0.0090

0.0110

0.1549 0.1679 0.1901

0.0040

0.1891 0.0078 0.0042

0.1541 0.1676

BER

NC

NC

BER

Using 7 repetition code

Using 5 repetition code

Scaling(zoomout=0.5,zoomin=2) Scaling(zoomout=0.25,zoomin=4)

Lena

Attack

0.9995

0.9991 0.9991

0.8806 0.8826

0.8992

0.9950

0.9937

0.9932

0.8838 0.8761 0.8532

0.9991

0.8546 0.9995 0.9982

0.8843 0.8767

NC

0.0004

0.0015 0.0004

0.1402 0.1314

0.1302

0.0046

0.0049

0.0056

0.1361 0.1429 0.1614

0.0012

0.1602 0.0020 0.0024

0.1358 0.1424

BER

Using 9 repetition code

1

1 1

0.9182 0.9021

0.9189

0.9959

0.9959

0.9968

0.9029 0.8981 0.8751

1

0.8760 0.9991 0.9991

0.9059 0.8993

NC

0.0004

0.0004 0.0004

0.0985 0.1072

0.0991

0.0032

0.0029

0.0027

0.1118 0.1135 0.1471

0.0004

0.1461 0.0020 0.0004

0.1109 0.1127

BER

Using 11 repetition code

Table 9 NC and BER comparison of extracted watermark using 5 Repetition Code,7 Repetition Code,9 Repetition Code and 11 Repetition Code under scaling attacks

Multimed Tools Appl

0.8561 0.8426 0.8086

0.9928

0.9950 0.9951

Scaling(zoomout=2,zoomin=0.5)

Scaling(zoomout=4,zoomin=0.25) Scaling(zoomout=8,zoomin=0.125)

0.0063 0.0059

0.0095

0.1609 0.1758 0.2197 0.9986 0.9991

0.9973

0.8678 0.8587 0.8309 0.0012 0.0015

0.0022

0.1538 0.1609 0.1906

BER

NC

NC

BER

Using 7 repetition code

Using 5 repetition code

Scaling(zoomout=0.5,zoomin=2) Scaling(zoomout=0.25,zoomin=4) Scaling(zoomout=0.125,zoomin=8)

Pepper

Attack

Table 9 (continued)

0.9986 0.9986

0.9991

0.8906 0.8782 0.8628

NC

0.0004 0.0004

0.0015

0.1341 0.1418 0.1608

BER

Using 9 repetition code

1 1

1

0.9152 0.9018 0.8784

NC

0.0004 0

0.0004

0.1018 0.1113 0.1443

BER

Using 11 repetition code

Multimed Tools Appl

Multimed Tools Appl PSNR comparison of pepper image under scaling aack

45 40 35 30 25 20 15 10 5 0

USING 5 REPETITION CODE

USING 7 REPETITION CODE

USING 9 REPETITION CODE

USING 11 REPETITION CODE

Fig. 12 Imperceptibility verification of proposed scheme under scaling attacks

and BER values of recovering watermark are put on the Table 9 for analogous test images. Recover watermarks of scaling attacks are exclusively represented the watermark logo. This scheme is imperceptible against scaling attacks as demonstrated in Fig. 12. (xiii) Deletion of lines or columns: Very frequently some part of the image is removed or deleted intentionally or unintentionally. This is one of the geometric transformation attack that have been widely used to the some simple copyright protection watermarking system. Sometimes this deletion of lines or columns has the same effect as image scaling. The proposed technique is verified against this type of attack for better robustness and improved invisibility. In this experiment, some lines (rows) or columns of watermarked images are deleted. If deletion in row wise, deletes some upper rows and lower rows where row numbers are same. For column deletion, delete same number of columns from left and right side of the watermarked images as one 40 column deleted watermarked image is shown in Fig. 14o. PSNR comparison diagram in Fig. 13 explains the invisibility property of the proposed scheme against this attack. NC and BER values of recover watermark under the different row and column sizes are shown in Table 10 for corresponding cover images. NC and BER values of recovering logo prove the consistency of the proposed scheme under different row and column deletion attacks along with different repetition codes. This is presented in matching recovered watermark logo table (Table 10). (xiv) Translation: Sometimes deletion of subpart of an image and geometric transformation of the rest part of the an image is done in such a way that it looks like the image is translated into a new coordinate. To provide better performance against translation attack proposed technique is verified against it by varying translation coordinate. In Fig. 14p, one translation attack image with translation coordinate (10,10) is presented PSNR comparison of pepper image under various aack

20 10 0 Cut(20 rows)

Cut(20 columns)

USING 5 REPETITION CODE

Cut(30 rows)

Cu t(30 columns)

USING 7 REPETITION CODE

Cut(40 rows)

Cut(40 columns)

USING 9 REPETITION CODE

Image Translation(10,10)

Image Shearing(0.05,0.05)

USING 11 REPETITION CODE

Fig. 13 Imperceptibility verification of proposed scheme under various geometric attacks

Multimed Tools Appl Table 10 NC and BER comparison of extracted watermark using 5 Repetition Code,7 Repetition Code,9 Repetition Code and 11 Repetition Code under various attacks

Attack

Lena Cut(20 rows)

Using 5 repetition code

Using 7 repetition code

Using 9 repetition code

Using 11 repetition code

NC

NC

NC

NC

BER

BER

BER

BER

0.9891 0.0146 0.9991 0.0059 0.9995 0.0076 0.9997 0.0061

Cut(20 columns) Cut(30 rows) Cut(30 columns)

0.9882 0.0149 0.9964 0.0059 0.9973 0.0046 0.9974 0.0042 0.9829 0.0154 0.9888 0.0088 0.9914 0.0100 0.9936 0.0095 0.9816 0.0176 0.9874 0.0110 0.9879 0.0112 0.9914 0.0083

Cut(40 rows) Cut(40 columns)

0.9878 0.0137 0.9968 0.0051 0.9977 0.0042 0.9979 0.0039 0.9869 0.0127 0.9950 0.0054 0.9973 0.0037 0.9978 0.0034

Image Translation(10,10) 0.8625 0.1632 0.8752 0.1436 0.9036 0.1095 0.9206 0.0904 Image Shearing(−0.05,0.05) 0.7914 0.2390 0.8105 0.2194 0.8409 0.1767 0.8741 0.1482 Airplane Cut(20 rows) Cut(20 columns)

0.9868 0.0178 0.9909 0.0129 0.9945 0.0107 0.9964 0.0076 0.9864 0.0176 0.9869 0.0127 0.9919 0.0076 0.9946 0.0054

Cut(30 rows)

0.9820 0.0183 0.9825 0.0151 0.9869 0.0137 0.9905 0.0132

Cut(30 columns) Cut(40 rows) Cut(40 columns)

0.9767 0.0225 0.9759 0.0208 0.9799 0.0156 0.9847 0.0139 0.9851 0.0166 0.9870 0.0120 0.9905 0.0085 0.9928 0.0071 0.9855 0.0159 0.9861 0.0122 0.9906 0.0083 0.9919 0.0076

Image Translation(10,10) 0.8617 0.1629 0.8746 0.1431 0.9031 0.1092 0.9201 0.0901 Image Shearing(−0.05,0.05) 0.7905 0.2395 0.8094 0.2181 0.8385 0.1756 0.8732 0.1471 Barbara Cut(20 rows) Cut(20 columns) Cut(30 rows) Cut(30 columns) Cut(40 rows) Cut(40 columns) Image Translation(10,10)+ Image Shearing(−0.05,0.05) Pepper Cut(20 rows)

0.9927 0.9923 0.9868 0.9812 0.9923

0.0127 0.0100 0.0117 0.0164 0.0078

0.9982 0.9946 0.9874 0.9839 0.9964

0.0073 0.0061 0.0105 0.0125 0.0042

0.9991 0.9955 0.9892 0.9865 0.9973

0.0105 0.0059 0.0122 0.0117 0.0039

1 0.9977 0.9932 0.9914 0.9981

0.0066 0.0042 0.0093 0.0090 0.0039

0.9932 0.0068 0.9955 0.0046 0.9964 0.0042 0.9982 0.0034 0.8662 0.1614 0.8778 0.1424 0.9101 0.0998 0.9289 0.0794 0.7965 0.2374 0.8187 0.2162 0.8456 0.1701 0.8814 0.1301 0.9954 0.0105 1

0.0051 1

0.0078 1

0.0063

Cut(20 columns) Cut(30 rows)

0.9946 0.0088 0.9955 0.0051 0.9955 0.0044 0.9941 0.0051 0.9923 0.0076 0.9924 0.0059 0.9919 0.0076 0.9941 0.0066

Cut(30 columns) Cut(40 rows) Cut(40 columns) Image Translation(10,10) Image Shearing(−0.05,0.05)

0.9860 0.9964 0.9965 0.8651 0.7958

0.0139 0.0061 0.0059 0.1634 0.2379

0.9866 0.9986 0.9968 0.8785 0.8189

0.0103 0.0024 0.0032 0.1429 0.2161

0.9896 0.9986 0.9977 0.9121 0.8516

0.0105 0.0029 0.0034 0.0991 0.1682

0.9928 0.9986 0.9986 0.9291 0.8918

0.0085 0.0032 0.0032 0.0792 0.1287

Multimed Tools Appl

Fig. 14 Result under different types of attack and recovered watermark images using (i) 5 Repetition code (ii) 7 Repetition code (iii) 9 Repetition code (iv) 11 Repetition code a Gaussian Lowpass Filter(2,2) b Median Filter(2,2) c Wiener Filter(2,2) d Average Filter(2,2) e Image Sharpening f Histogram Equalization g Gamma correction with gamma=3 h Gamma attack(gamma=0.25) i Salt & Peppers noise(density=0.01) j Speckle noise (var=0.01) k Gaussian noise (M=0,var=0.01) l Rotation anticlockwise 5° m Cropping (256 × 256 by black) n Image Scaling (512->64->512) o Cut (40 columns) p Translation (10,10) q Image Shearing(−0.05,0.05) r JPEG compression Q=50 s Gamma attack (gamma=3) + Image Shearing t Cropping (128 × 128 by white) + Rotation clockwise 5°

as well as corresponding recovered watermark is shown for four different repetition codes. After applying different translation value to four test watermarked images, NC and BER values of recovering watermark are estimated in related Table 10. These NC and BER values explicitly confirms the robustness of the proposed scheme. Chart in Fig. 13 concludes the imperceptibility property of the scheme in terms of PSNR values. (xv) Shearing: To confirm the robustness against shear operation, the proposed technique is tested against image shearing with different parameter. Actually, shearing of an image means distortion of the image in the x-direction or y-direction or in both the direction. Here one watermarked image is sheared in both directions with different parameter values is shown in Fig. 14q. BER and NC values of recovered watermark for different test images are tabulated in associated Table 10. Uniform invisibility of the proposed method against shearing attack is given in Fig. 13. To prove robustness consistency under shearing operation, BER and NC values of recovering watermark images are tabulated in the above table by changing repetition code values.

Multimed Tools Appl Table 11 NC and BER comparison of extracted watermark using 5 Repetition Code,7 Repetition Code,9 Repetition Code and 11 Repetition Code under various attacks

Attack

Lena JPEG compression Q=40 JPEG compression Q=50 JPEG compression Q=60 JPEG compression Q=70 Gamma attack(gamma=3) + Image Shearing Cropping(128 × 128 by white) + Rotation clockwise 5 degree JPEG compression Q=50 + Cropping(128 × 128 by white) Airplane JPEG compression Q=40 JPEG compression Q=50 JPEG compression Q=60 JPEG compression Q=70 Gamma attack(gamma=3) + Image Shearing Cropping(128 × 128 by white) + Rotation clockwise 5 degree JPEG compression Q=50 + Cropping(128 × 128 by white) Barbara JPEG compression Q=40 JPEG compression Q=50 JPEG compression Q=60 JPEG compression Q=70 Gamma attack(gamma=3) + Image Shearing Cropping(128 × 128 by white) + Rotation clockwise 5 degree JPEG compression Q=50 + Cropping(128 × 128 by white) Pepper JPEG compression Q=40 JPEG compression Q=50 JPEG compression Q=60 JPEG compression Q=70 Gamma attack(gamma=3) + Image Shearing Cropping(128 × 128 by white) + Rotation clockwise 5 degree JPEG compression Q=50 + Cropping(128 × 128 by white)

Using 5 Using 7 Using 9 Using 11 repetition code repetition code repetition code repetition code NC

BER

NC

BER

NC

BER

NC

BER

0.9919 0.9920 0.9927 0.9931 0.7462

0.0078 0.0071 0.0068 0.0067 0.2837

0.9947 0.9986 0.9991 0.9992 0.7642

0.0049 0.0022 0.0012 0.0012 0.2635

0.9964 0.9986 0.9994 1 0.7941

0.0022 0.0012 0.0008 0.0001 0.2401

0.9986 0.9995 0.9995 1 0.8131

0.0015 0.0004 0.0004 0.0001 0.2165

0.7921 0.2387 0.8113 0.2198 0.8335 0.1899 0.8641 0.1602 0.7856 0.2381 0.8067 0.2201 0.8229 0.2036 0.8469 0.1729

0.9902 0.9914 0.9921 0.9923 0.7458

0.0083 0.0073 0.0070 0.0070 0.2839

0.9969 0.9981 0.9989 0.9992 0.7635

0.0029 0.0024 0.0013 0.0012 0.2639

0.9958 0.9982 0.9993 1 0.7935

0.0030 0.0017 0.0008 0.0001 0.2417

0.9981 0.9995 0.9995 1 0.8126

0.0018 0.0004 0.0004 0.0001 0.2171

0.7912 0.2381 0.8105 0.2191 0.8327 0.1892 0.8638 0.1598 0.7842 0.2389 0.8058 0.2197 0.8221 0.2041 0.8467 0.1730

0.9899 0.9909 0.9922 0.9932 0.7451

0.0084 0.0075 0.0069 0.0065 0.2841

0.9962 0.9979 0.9987 0.9994 0.7638

0.0030 0.0025 0.0014 0.0010 0.2637

0.9968 0.9991 0.9994 1 0.8016

0.0021 0.0012 0.0006 0.0001 0.2365

0.9983 0.9995 1 1 0.8212

0.0017 0.0004 0.0001 0.0001 0.2158

0.7902 0.2387 0.8154 0.2174 0.8398 0.1804 0.8702 0.1542 0.7812 0.2397 0.8051 0.2199 0.8356 0.2078 0.8489 0.1718

0.9908 0.9911 0.9931 0.9939 0.8658

0.0080 0.0073 0.0061 0.0051 0.1619

0.9971 0.9980 0.9991 0.9993 0.8785

0.0028 0.0024 0.0012 0.0010 0.1419

0.9986 0.9991 1 1 0.9098

0.0019 0.0012 0.0004 0.0001 0.0999

0.9992 1 1 1 0.9291

0.0010 0.0004 0.0001 0.0001 0.0784

0.7432 0.2864 0.7659 0.2629 0.8119 0.2309 0.8318 0.2116 0.7897 0.2392 0.8186 0.2148 0.8562 0.1636 0.8746 0.1506

Multimed Tools Appl

4.4 Compression attack (xvi)

JPEG compression: JPEG compression is one of the most widely used common manipulation/compression attack. Any effective watermark should be rigid to some degree of compression. Proposed scheme is checked against JPEG compression by changing quality factor values. Recovered logos NC and BER values from the compressed watermarked image under different quality factor are put into the equivalent Table 11. Results from above table provide enough data to confirm the improvement of recovering watermark quality with the increment of JPEG compression quality factor and with the enhancement of repetition code values. Reliable imperceptibility against compression attack is plotted in Fig. 15 (in terms of PSNR values of watermarked images). One compression attack image (quality factor=50) along with recovered logos using 5 repetition code,7 repetition code,9 repetition code,11 repetition code are given in Fig. 14r and it proves the robustness property of the proposed scheme against compression attack .

4.5 Combinational attack For robustness test, some combination of above discussed attacks is applied to all the standard test images. There can be lots of combinational attacks, but here three different combinational attacks are presented as follows: (xvii) Combination of one enhancement technique attack and one geometric attack: For robustness clarity, combinational attacks of one enhancement technique attack and one geometric attack are applied on watermarked images of the proposed scheme. From the common image enhancement technique, gamma attack (gamma=3) technique along with shearing geometric transformation attack have been selected for combinational attack and the attack image of this combinational attack is portrayed in Fig. 14s. Recovered watermarks for this attack using 5 repetition code,7 repetition code,9 repetition code,11 repetition code are also given in same figure. Watermarked images PSNR values under this attack prove the constancy against invisibility measurement. This information is given in Fig. 15. Recovered logos NC and BER values for all the test images explain the watermark recover capability of the proposed scheme for this type of combinational attack. These values are put in the Table 11. (xx) Combination of geometric attack with geometric attack: One geometric attack alone may not be enough to prove the robustness of the proposed scheme. As a consequence, watermarked images of this method are tested against the combination of one geometric PSNR comparison of pepper image under various aack

40 20 0 JPEG compression Q=40 JPEG compression Q=50 JPEG compression Q=60 JPEG compression Q=70

USING 5 REPETITION CODE

USING 7 REPETITION CODE

Cropping(128 × 128 by JPEG compression Q=50 white) + Rotaon + Cropping(128 × 128 by white) clockwise 5 degree

Gamma aack(gamma=3) + Image Shearing

USING 9 REPETITION CODE

USING 11 REPETITION CODE

Fig. 15 Imperceptibility verification of proposed scheme under various attacks

Multimed Tools Appl Table 12 NC and BER comparison of extracted watermark using 5 Repetition Code,7 Repetition Code,9 Repetition Code and 11 Repetition Code under various attacks for BCheckmark^ attacks

BCheckmark^ Attack

Using 5 repetition code

Using 7 repetition code

Using 9 repetition code

Using 11 repetition code

NC

NC

NC

NC

BER

BER

BER

BER

Median Filter(2,2)

0.8492 0.1662 0.8584 0.1615 0.9102 0.0939 0.9386 0.0627

Median Filter(3,3) Wiener Filter (2,2) Wiener Filter (3,3) Hard Threshold Soft Threshold

0.7325 0.8982 0.7425 0.8217 0.7849

Image Sharpening

0.9810 0.0198 0.9921 0.0085 0.9945 0.0079 0.9972 0.0045

0.2914 0.1164 0.2810 0.2062 0.2514

0.7351 0.9131 0.7489 0.8279 0.7943

0.2889 0.0978 0.2795 0.1996 0.2364

0.7401 0.9529 0.7541 0.8406 0.8174

0.2853 0.0480 0.2776 0.1797 0.2148

0.7618 0.9669 0.7847 0.8606 0.8362

0.2721 0.0329 0.2443 0.1614 0.1841

Gaussian noise(M=0, var=0.005) 0.8072 0.2216 0.8218 0.2057 0.8427 0.1892 0.8527 0.1621 Rotation(clockwise 5°) 0.8402 0.1799 0.8682 0.1542 0.8862 0.1347 0.8915 0.1197 Cropping(64 × 64 by white) 0.9887 0.0112 0.9890 0.0109 0.9945 0.0074 0.9991 0.0027 Scaling(factor=2.5) Cut(20 rows)/Rows Removal

0.9814 0.0221 0.9923 0.0097 0.9940 0.0069 0.9981 0.0034 0.9875 0.0164 0.9927 0.0091 0.9956 0.0051 0.9972 0.0040

Cut(20 columns)/Columns Removal

0.9864 0.0161 0.9951 0.0067 0.9961 0.0047 0.9965 0.0042

Template Removal Image Shearing(−0.05,0.05)

0.7583 0.2697 0.7693 0.2691 0.7834 0.2529 0.8084 0.2198 0.7897 0.2405 0.8082 0.2218 0.8378 0.1804 0.8717 0.1509

Warping 0.7652 0.2647 0.7781 0.2608 0.7938 0.2347 0.8197 0.2143 JPEG compression Q=50 0.9895 0.0097 0.9965 0.0043 0.9974 0.0041 0.9987 0.0021 Rotation(clockwise 5°) + Scaling 0.7438 0.2819 0.7541 0.2734 0.7603 0.2708 0.7946 0.2351 Dithering Remodulation Downsampling/Upsampling

0.7283 0.3152 0.7331 0.2954 0.7425 0.2816 0.7527 0.2772 0.7512 0.2774 0.7641 0.2668 0.7843 0.2469 0.8096 0.2189 0.7387 0.2918 0.7512 0.2790 0.7683 0.2619 0.7857 0.2452

transformation attack with another type of geometric transformation attack. In Fig. 14t, combinational attack image by cropping (128 × 128 by white) and rotation (clockwise 5°) is presented along with recovered logos using four different types of repetition Comparison of proposed scheme with some exisng schemes interms of revovered logo's NC value 1.5 1 0.5 0 A.Gaussian Filter

B.Median filter

C.Image sharpening

D.Rotaon (5°)

E.Cropping

F.JPEG(QF=50)

G.JPEG(QF=70)

H.Scaling

USING 5 REPETITION CODE

USING 7 REPETITION CODE

USING 9 REPETITION CODE

USING 11 REPETITION CODE

Wang et al.[25]

Lien et al.[26]

Lin et al.[27]

Das et al.[11]

Fig. 16 Comparative analysis of proposed scheme with some existing schemes

I.Combinaonal aack

0.9928

0.9980 0.9993 0.8678

0.8186

0.9882 0.9911 0.9939 0.8561

F. JPEG(Q=50) G. JPEG(Q=70) H. Scaling

I. Combinational attack 0.7897

E. Cropping

0.8602 0.9989 0.8887

0.8596 0.9963 0.8617

0.9986

0.9964

B. Median filter C. Image sharpening D. Rotation(5°)

0.8562

0.9991 1 0.8906

0.9959

0.9271 0.9994 0.9120

0.9995

0.8746

1 1 0.9152

0.9969

0.9509 1 0.9312

1

5 repetition code 7 repetition code 9 repetition code 11 repetition code

Proposed scheme

A. Gaussian filter

Attack

0.89 0.97 0.79 –

–

0.92

0.79 0.06 0.53

0.84

0.28 0.57 –

-

0.51 0.46 0.37

0.64

–

0.96 0.97 0.88

0.66

0.92 0.97 0.59

0.92

0.8641

0.9810 0.9918 –

0.9954

0.9118 0.9327 0.8762

0.9118

Wang and Lin [52] Lien and Lin [24] Lin et al. [25] Das et al. [11]

Table 13 Comparative study of the proposed scheme with some existing schemes in terms of recovered logo’s NC value

Multimed Tools Appl

Multimed Tools Appl

codes. NC and BER values of recovered logos for all the standard test images are presented in correlated Table 11 whereas the stable imperceptibility property of the proposed method is graphically represented in Fig. 15. (xix) Combination of one compression attack and one geometric attack: For robustness confirmation, watermarked images of this scheme are put against combination of one JPEG compression attack and one geometric attack. In this experiment one JPEG compression (quality factor=50) and Cropping (128 × 128 by white) are opted for these types of combinational attack. NC and BER values of recovered logos are tabulated in the related Table 11 after applying above mentioned combinational attack. Experimental data rows in above mentioned table validate the robustness of this method using different repetition codes. PSNR chart in Fig. 15 describes the invisibility property of the proposed technique.

4.5.1 Comparative analysis The proposed scheme is compared with some existing watermarking schemes as shown in Table 12. The experimental results in Table 12 and graphical analysis in Fig. 16 conclude the superiority of the proposed scheme with some existing schemes. Superior NC values of recovering watermark logo under various attacks are shown as bold data in Table 13.

5 Conclusions In this paper, the proposed RDWT-DCT and repetition code based image watermarking scheme shows imperceptibility property. As RDWT is shift invariant, better PSNR and high robustness are beneficial point of the proposed hybrid method. One of the main objectives of the proposed blind watermarking method is to increase the watermark embedding capacity. Maximum embedding capacity of proposed technique is described in experimental result section. The incorporation of the repetition code in the proposed scheme increase robustness by resisting several image processing attacks. The watermark is scrambled in the preprocessing stage and distributed in all space of the host image at the time of embedding. This jumbling process of watermark not only provides extra security, but also improves the robustness of the proposed method. Furthermore, in the proposed scheme insertion of the chaotic watermark in the DCT transformed block’s selected AC coefficient pairs effectively resist the common image manipulation attacks like image enhancement by lowpass filtering, median filtering, wiener filtering, average filtering, image sharpening, histogram modification, gamma correction attack, image noising, geometric operations like image rotation, cropping, scaling, translation, shearing, deletion of lines or columns, JPEG compression attacks. In addition to this, the proposed scheme is verified against some combinational attacks of the previously described attacks. For better clarification, robustness performance of proposed scheme is verified with standard benchmark software BCheckmark^. Experimental results prove the supremacy of the proposed method with some existing schemes. The proposed scheme gives satisfactory result for grayscale image. However, in this proposed hybrid domain technique, computational complexity is on the higher side because of transform domain techniques like RDWT and DCT. Also, application of Arnold scrambling to the watermark in pre-processing stage further increase the computational complexity of this algorithm. It can be accepted because the main objective of this technique to improve the robustness and security of watermark. In conventional error correcting code (ECC) based

Multimed Tools Appl

watermarking schemes; redundant bits are added to the host image for error detection or error correction on the receiver side. In this proposed method, basic ECC repetition code is used to preserve one watermark bit in every decomposed non-overlapping host image block. So, implementation of proposed repetition code based watermarking scheme is less complex and computational complexity is also minimum than other transform domain ECC based watermarking methods. In future scope to reduce the computational complexity, single transform domain based blind watermarking technique can be devised with the help of repetition code and Arnold scrambling. In future we aim to improve the performance of proposed technique in terms of robustness, imperceptibility, embedding capacity, computational complexity and security requirements with the help of other error correcting codes (like Hamming, the Bose, Ray-Chaudhuri, Hocquenghem (BCH), the Reed–Solomon and some hybrid error correcting codes). Also, we want to carry the same work for devising effective color image watermarking scheme as watermark embedding space is increased in color image watermarking. It can improve the performance of the color image watermarking technique. But selection of red/green/blue color channel gives better performance or YCbCr based color image technique gives better performance that will be examined in future. In addition to this, our future work will focus on developing readily embeddable and extractable watermarking method for real-time applications.

References 1. BCheckmark^ Available at http://watermarking.unige.ch/checkmark/ 2. Ali M, Ahn CW, Pant M (2014) A robust image watermarking technique using SVD and differential evolution in DCT domain. Optik-Int J Light Electron Opt 125(1):428–434 3. Barni M, Bartolini F, Cappellini V, Piva A (1998) A DCT-domain system for robust image watermarking. Signal Processing 66(3):357–372 4. Bhandari K, Mitra SK, JadhavA (2005) A hybrid approach to digital image watermarking using singular value decomposition and spread spectrum. In: Pattern Recognition and Machine Intelligence: 447–452 5. Campisi P, Kundur D, Neri A (2004) Robust digital watermarking in the ridgelet domain. IEEE Signal Process Lett 11(10):826–830 6. Celik MU, Sharma G, Tekalp AM, Saber E (2005) Lossless generalized-LSB data embedding. IEEE Trans Image Process 14(2):253–266 7. Chang CC, Tsai P, Lin CC (2005) SVD-based digital image watermarking scheme. Pattern Recogn Lett 26(10):1577–1586 8. Chu WC (2003) DCT-based image watermarking using subsampling. IEEE Trans Multimedia 5(1):34–38 9. Chung KL, Shen CH, Chang LC (2001) A novel SVD-and VQ-based image hiding scheme. Pattern Recogn Lett 22(9):1051–1058 10. Chung KL, Yang WN, Huang YH, Wu ST, Hsu YC (2007) On SVD-based watermarking algorithm. Appl Math Comput 188(1):54–57 11. Das C, Panigrahi S, Sharma VK, Mahapatra KK (2014) A novel blind robust image watermarking in DCT domain using inter-block coefficient correlation. AEU Int J Electron Commun 68(3):244–253 12. Deng F, Wang B (2003) A novel technique for robust image watermarking in the DCT domain. In: Neural Networks and Signal Processing. IEEE Conference Proceedings 2:1525–1528 13. Fan MQ, Wang HX, Li SK (2008) Restudy on SVD-based watermarking scheme. Appl Math Comput 203(2):926–930 14. Fu Y (2013) Robust oblivious image watermarking scheme based on coefficient relation. Optik-Int J Light Electron Opt 124(6):517–521 15. Ganic E, Eskicioglu AM (2004) Robust DWT-SVD domain image watermarking: embedding data in all frequencies. In: Proceedings of the ACM Workshop on Multimedia and Security:166–174 16. Ghazy RA, El-fishawy NA, Hadhoud MM, Dessouky MI, El-Samie FE (2007) An efficient block-by-block SVD-based image watermarking scheme. In: IEEE Radio Science Conference, NRSC 2007:1–9 17. Gravano, S. (2001). Introduction to error control codes (Vol. 9). Oxford University Press on Demand. 18. Kundur D, Hatzinakos D (1998) Digital watermarking using multiresolution wavelet decomposition. In: Acoustics, Speech and Signal Processing. IEEE Conference Proceedings 5:2969–2972

Multimed Tools Appl 19. Kutter M, Petitcolas FA (1999) Fair benchmark for image watermarking systems. In Electronic Imaging’99. International Society for Optics and Photonics:226–239 20. Lagzian S, Soryani M, Fathy M (2011) A new robust watermarking scheme based on RDWT-SVD. Int J Intell Inf Process 2(1):22–29 21. Langelaar GC, Van Der Lubbe JC, Biemond J (1996) Copy protection for multimedia data based on labeling techniques. In: Symposium on Information Theory in the Benelux 1996:33–40 TECHNISCHE UNIVERSITEIT DELFT 22. Liang J, Tran TD (2001) Fast multiplierless approximations of the DCT with the lifting scheme. IEEE Trans Signal Process 49(12):3032–3044 23. Liang X, Zhihui W, Huizhong W (2006) Ridgelet-based robust and perceptual watermarking for images. IJCSNS Int J Comput Sci Netw Secur 6(2):194–201 24. Lien BK, Lin WH (2006) A watermarking method based on maximum distance wavelet tree quantization. In: Proceedings 19th Conference Computer Vision, Graphics and Image Processing: 269–276 25. Lin WH, Wang YR, Horng SJ, Kao TW, Pan Y (2009) A blind watermarking method using maximum wavelet coefficient quantization. Expert Syst Appl 36(9):11509–11516 26. Lin CY, Wu M, Bloom JA, Cox IJ, Miller ML, Lui YM (2001) Rotation, scale, and translation resilient watermarking for images. IEEE Trans Image Process 10(5):767–782 27. Liu JC, Chen SY (2001) Fast two-layer image watermarking without referring to the original image and watermark. Image Vis Comput 19(14):1083–1097 28. Liu R, Tan T (2002) An SVD-based watermarking scheme for protecting rightful ownership. IEEE Trans Multimedia 4(1):121–128 29. Ma F, Zhang J, Zhang W (2012) A blind watermarking technology based on DCT domain. In: Computer Science & Service System (CSSS), IEEE International Conference on:397–400 30. Nikolaidis N, Pitas I (1996) Copyright protection of images using robust digital signatures. In: Acoustics, Speech, and Signal Processing, ICASSP-96. IEEE Conference Proceedings 4:2168–2171 31. Nikolaidis N, Pitas I (1998) Robust image watermarking in the spatial domain. Signal Process 66(3):385– 403 32. O’Ruanaidh JJ, Dowling WJ, Boland FM (1996) Phase watermarking of digital images. In: IEEE International Conference on Image Processing:239–242 33. Ouhsain M, Hamza AB (2009) Image watermarking scheme using nonnegative matrix factorization and wavelet transform. Expert Syst Appl 36(2):2123–2129 34. Pandey R, Singh AK, Kumar B, Mohan A (2016) Iris based secure NROI multiple eye image watermarking for teleophthalmology. Multimed Tools Appl:1–7 35. Pereira S, Voloshynovskiy S, Madueño M, Marchand-Maillet S, Pun T (2001) Second generation benchmarking and application oriented evaluation. In: Information hiding workshop III, Pittsburgh, PA, USA: 340–353 36. Phadikar A, Maity SP, Verma B (2011) Region based QIM digital watermarking scheme for image database in DCT domain. Comput Electr Eng 37(3):339–355 37. Premaratne P, Ko CC (1999) A novel watermark embedding and detection scheme for images in DFT domain. In: Image Processing and Its Applications. IET Seventh International Conference 2:780–783 38. Quan L, Qingsong A (2004) A combination of DCT-based and SVD-based watermarking scheme. In: Signal Processing, ICSP’04. IEEE Proceedings on 1:873–876 39. Rastegar S, Namazi F, Yaghmaie K, Aliabadian A (2011) Hybrid watermarking algorithm based on singular value decomposition and radon transform. AEU Int J Electron Commun 65(7):658–663 40. Raval MS, Rege PP (2003) Discrete wavelet transform based multiple watermarking scheme. In: IEEE Conference on Convergent Technologies 3:935–938 41. Singh AK (2016) Improved hybrid algorithm for robust and imperceptible multiple watermarking using digital images. Multimed Tools Appl:1–8 42. Singh AK, Dave M, Mohan A (2014) Hybrid technique for robust and imperceptible image watermarking in DWT–DCT–SVD domain. Natl Acad Sci Lett 37(4):351–358 43. Singh AK, Dave M, Mohan A (2015) Robust and secure multiple watermarking in wavelet domain. J Med Imaging Health Inform 5(2):406–414 44. Singh AK, Dave M, Mohan A (2015) Hybrid technique for robust and imperceptible multiple watermarking using medical images. Multimed Tools Appl:1–21

Multimed Tools Appl 45. Singh AK, Kumar B, Dave M, Mohan A (2015) Multiple watermarking on medical images using selective discrete wavelet transform coefficients. J Med Imaging Health Inform 5(3):607–614 46. Singh AK, Kumar B, Dave M, Mohan A (2015) Robust and imperceptible dual watermarking for telemedicine applications. Wirel Pers Commun 80(4):1415–1433 47. Solachidis V, Pitas I (2001) Circularly symmetric watermark embedding in 2-D DFT domain. IEEE Trans Image Process 10(11):1741–1753 48. Swanson MD, Zhu B, Tewfik AH (1996) Transparent robust image watermarking. In: IEEE Image Processing Proceedings 3:211–214 49. Tay P, Havlicek JP (2002) Image watermarking using wavelets. In: Circuits and Systems, MWSCAS-2002. IEEE Midwest Symposium 3:258–261 50. Veeraswamy K, Kumar SS (2008) Adaptive AC-coefficient prediction for image compression and blind watermarking. J Multimed 3(1):16–22 51. Voloshynovskiy S, Pereira S, Pun T, Eggers JJ, Su JK (2001) Attacks on digital watermarks: classification, estimation based attacks, and benchmarks. IEEE Commun Mag 39(8):118–126 52. Wang SH, Lin YP (2004) Wavelet tree quantization for copyright protection watermarking. IEEE Trans Image Process 13(2):154–165 53. Wang Y, Pearmain A (2004) Blind image data hiding based on self reference. Pattern Recogn Lett 25(15): 1681–1689 54. Ward DL (2003) Redundant discrete wavelet transform based super-resolution using sub-pixel image registration. Air Force Institute of Technology 55. Wu C, Zhu WP, Swamy MN (2004) A watermark embedding scheme in wavelet transform domain. In: TENCON, IEEE Region 10 Conference Proceedings A:279–282 56. Yantao Z, Yunfei M, Zhiquan L (2008) A robust chaos-based DCT-domain watermarking algorithm. In: Computer Science and Software Engineering. IEEE International Conference on 3:935– 938

Soumitra Roy is working as an Assistant Professor in the Department of Computer Science and Engineering, Dr. B.C. Roy Engineering College Durgapur, India. He received his BTech in CSE and MTech in CSE from Maulana Abul Kalam Azad University of Technology, West Bengal (formerly known as West Bengal University of Technology). He has around seven years of teaching and research experiences and his research interest includes digital image processing, steganography and watermarking.

Multimed Tools Appl

Arup Kumar Pal is presently working as an Assistant Professor in the Department of Computer Science and Engineering, Indian School of Mines, Dhanbad, India. Prior to join this institute, he was a Lecturer in the Department of Computer Science and Engineering, NIT Jamshedpur during April, 2011 to December, 2011. He did his PhD in Computer Science and Engineering from Indian School of Mines, Dhanbad in 2011. He has around five years of teaching and research experiences, and contributed a number of research papers in several journals and conference proceedings of National and International reputes. His main research interest includes vector quantization, image compression, image cryptosystem, steganography, watermarking and CBIR.