J. Cent. South Univ. (2017) 24: 521-532

DOI: 10.1007/s11771-017-3455-3

A new image watermarking framework based on levels-directions decomposition in contourlet representation

M. F. Kazemi¹, M. A. Pourmina¹, A. H. Mazinan²

1. Department of Electrical and Computer Engineering, Science and Research Branch,

Islamic Azad University (IAU), Tehran, Iran;

2. Department of Control Engineering, Faculty of Electrical Engineering, South Tehran Branch,Islamic Azad University (IAU), Tehran, Iran

Central South University Press and Springer-Verlag Berlin Heidelberg 2017

Abstract: With the development of digital information technologies, robust watermarking framework is taken into real consideration as a challenging issue in the area of image processing, due to the large applicabilities and its utilities in a number of academic and real environments. There are a wide range of solutions to provide image watermarking frameworks, while each one of them is attempted to address an efficient and applicable idea. In reality, the traditional techniques do not have sufficient merit to realize an accurate application. Due to the fact that the main idea behind the approach is organized based on contourlet representation, the only state-of-the-art materials that are investigated along with an integration of the aforementioned contourlet representation in line with watermarking framework are concentrated to be able to propose the novel and skilled technique. In a word, the main process of the proposed robust watermarking framework is organized to deal with both new embedding and de-embedding processes in the area of contourlet transform to generate watermarked image and the corresponding extracted logo image with high accuracy. In fact, the motivation of the approach is that the suggested complexity can be of novelty, which consists of the contourlet representation, the embedding and the corresponding de-embedding modules and the performance monitoring including an analysis of the watermarked image as well as the extracted logo image. There is also a scrambling module that is working in association with levels-directions decomposition in contourlet embedding mechanism, while a decision maker system is designed to deal with the appropriate number of sub-bands to be embedded in the presence of a series of simulated attacks. The required performance is tangibly considered through an integration of the peak signal-to-noise ratio and the structural similarity indices that are related to watermarked image. And the bit error rate and the normal correlation are considered that are related to the extracted logo analysis, as well. Subsequently, the outcomes are fully analyzed to be competitive with respect to the potential techniques in the image colour models including hue or tint in terms of their shade, saturation or amount of gray and their brightness via value or luminance and also hue, saturation and intensity representations, as long as the performance of the whole of channels are concentrated to be presented. The performance monitoring outcomes indicate that the proposed framework is of significance to be verified.

Key words: contourlet based watermarking framework; levels-directions decomposition; embedding process; de-embedding process; peak signal-to-noise ratio;structural similarity indices; normal correlation; bit error rate

1 Introduction

The present research attempts to address a new robust image watermarking framework utilizing contourlet representation. It is to note that the main process of watermarking framework is realized in the proposed research via the contourlet embedding technique, while a series of attacks are applied to the outcomes, in order to present an applicable and efficient approach, which can be useful in real environments. The idea of realizing the framework in complicated domains is completed, since a scrambling module is also employed to represent the information in disorder. In fact, it aims us to organize a complicated system to be able to process a set of separated images with different qualities, coherently. In a word, the approach is an integration of new embedding and de-embedding processes, as long as the ability of the contourlet representation is directly used to generate watermarked image and the corresponding extracted logo image with high accuracy. It is shown that the results are considered in a number of standard colour models including hue or tint in terms of their shade, saturation or amount of gray and their brightness via value or luminance (HSV) and hue, saturation and intensity (HSI), as long as all the channels performance are completely concentrated to be presented. There is a performance monitoring to evaluate both watermarked image and also the corresponding extracted logo image, as well.Prior to considering the contourlet based watermarking framework proposed, in detail, it is important to survey the recent potential investigations with respect to state-of-the-art, where the integration of watermarking framework and contourlet representation are exactly researched. With this, SADREAZAMI et al [1] study a multiplicative watermarking scheme in the field of contourlet domain. There is a model of the contourlet coefficients through the alpha-stable distributions. The results indicate that the present distribution fits the empirical data accurately than its formerly designed distributions. It is also shown that the bivariate alpha-stable distribution may capture the across scale dependencies in case of the contourlet coefficients. The resulting realization yields a significantly reduced- complexity detector to provide a desirable performance, which is much superior with respect to that of the detector in line with the best-fit alpha-stable distribution. CHALAMALA et al [2] suggest an image watermarking technique along with contourlet transform to be compared its performance with a wavelet transform based technique. A watermark bit is in fact embedded into one of the eigenvalues regarding a fixed size block of the sub-band coefficients through contourlet transform. IDRISSI et al [3] present a method of signal processing based on contourlet transform and maximum entropy that inserted into a digital document to provide an invisible watermark. There is a code against any attack that may affect the watermarked data. LI et al [4] illustrate an image watermarking scheme through an encrypted kinoform in the area of the hybrid nonsubsampled contourlet transform in association with the singular value decomposition. In the approach documented here, a watermark is firstly moved into an encrypted kinoform under fractional Fourier transform as well as Gerchberg-Saxton algorithm. Subsequently, the host image is decomposed based on nonsubsampled contourlet transform, while the low frequency sub-band is divided into nonoverlapping blocks. The singular value decomposition is applied to each one of blocks. The kinoform is in fact embedded into the low frequency nonsubsampled contourlet transform coefficients by amending the largest singular value of each block.

SADREAZAMI et al [5] address the blind watermark detection problem, while the ability of the contourlet is concentrated. There is the contourlet coefficients of images that have non-Gaussian property to be represented by non-Gaussian distributions such as the normal inverse Gaussian. DUAN et al [6] propose a semi-fragile algorithm that is incorporated with the Hu moment invariants in the area of the wavelet-based contourlet transform. The invariance of parent as well as its children coefficients relations is used to deal with watermarks embedding, as long as the property of Hu moment invariants technique is adopted to classify the malicious and non-malicious regions in case of manipulations, once the suspected image is authenticated. SADREAZAMI et al [7] consider an image modeling in the area of contourlet domain, while the magnitudes of the coefficients are represented through a symmetric alpha- stable distribution, which is best realized to represent the transform coefficients along with a high non-Gaussian property as well as heavy tails. It is analyzed that the alpha-stable family regarding the distributions can provide a more accurate model to present the contourlet sub-band coefficients with respect to its formerly distributions as the generalized Gaussian and Laplacian distributions.

DUAN et al [8] present a semi-fragile algorithm, which is incorporated in the generalized Benford’s law in the area of wavelet-based contourlet transform. There is an adaptive threshold instead of a fixed one to improve the accuracy of the tampering region detection. It is to note that the adaptive process is designed along with the compression ratio of JPEG2000. The generalized Benford’s law is utilized, as long as the threshold is adapted. The tampering detection rate is in fact taken to evaluate the performance of the approach. RANJBAR et al [9] consider a blind robust watermarking technique consisting of two embedding stages, which the first one is to divide the odd description of image into non- overlapped fixed size blocks in line with the signature to be embedded in the high frequency component regarding the contourlet transform of the blocks. In the second one, the signature is embedded in the low frequency component regarding the global contourlet transform of the image. NIU et al [10] reports a color image watermarking algorithm with visual quality and reasonable resistance toward geometric distortions. There is the geometrically invariant space that is constructed through color image normalization, where a significant region is acquired from the normalized color image by using the invariant centroid theory. After that, the contourlet transform is carried out in the green channel of the significant region. In reality, the digital watermark is finally embedded into host color image by amending the low frequency contourlet transform coefficients. WANG et al [11] research a robust image watermarking based on exponent moments invariants in nonsubsampled contourlet transform domain. At first, the nonsubsampled contourlet transform domain is applied to host image and then the exponent moments of the low-pass nonsubsampled contourlet transform are computed to choose its accurate ones. The digital watermark is finally embedded by quantizing the modulus of the selected exponent moments.

SONG et al [12] propose a contourlet based image adaptive watermarking scheme, in which the watermark can be embedded into the contourlet coefficients in case of the image largest details. There is a robust scheme against the image processing attacks. The corresponding detection algorithm is presented to decide whether the watermark can be presented by exploiting the unique transform structure or not? It is shown that the proposed scheme is realized to detect by computing the correlation between the spread watermark and the corresponding watermarked image in the whole of contourlet sub-bands. SONG et al [13] present a robust digital image watermarking algorithm based on chaotic system and QR factorization. It is shown that the host images have to first be divided into blocks with the same size, where the present factorization is carried out on each one of blocks. In this case, pseudorandom circular chain is generated by logistic mapping, applied to select the embedding blocks to enhance the security of the scheme. There is column coefficients regarding the chosen blocks that are modified to embed watermarks without causing noticeable points. It should be noted that watermark extraction procedure can perform without the original cover image.

RAHIMI et al [14] suggest a dual and oblivious watermarking scheme in the area of contourlet domain. Due to the fact that the importance of region of interest in interpretation by medical doctors is of significance along with region of non-interest, the present adaptive dual watermarking scheme is suggested with different embedding strength in regions of interest and non- interest. There is embed watermark bits in singular value vectors for the embedded blocks under low pass sub-band in the area of contourlet domain. WANG et al [15] demonstrate an image watermarking scheme based on contourlet transform. The original carrier image is first executed in the area of contourlet transform, as long as four directions of the second level sub-band are marked. And the scrambled digital watermarking is embedded in them.

To introduce the proposed approach, in its brief form, it should be noted that the ultimate goal is to make a new contribution in the field of image watermarking framework through contourlet representation with respect to those obtained from the state-of-the-art outcomes considered here. The required number of modules to be designated are scrambling module, levels- directions decomposition, contourlet transform embedding and de-embedding processes, colour models, decision maker system and finally simulated attacks. It is to note that the main process of the proposed framework is organized to deal with the new embedding technique in the area contourlet transform to generate watermarked image, as long as the corresponding de-embedding technique is investigated to retrieve the watermarked image as well as the extracted logo image. The

motivation of the approach and the outcomes, which makes the technique distinct from each other potential ones are that the proposed structure is consisted of a new integration of a number of modules that are working to deal with high accuracy outcomes, which can be reported through an investigated performance monitoring module. It consists of the peak signal-to-noise ratio (PSNR) and the structural similarity indices (SSIM) for the watermarked image and also the bit error rate (BER) and the normal correlation (NC) for the extracted logo image, respectively. It aims us to evaluate the outcomes that are fully competitive with respect to the state-of-the-art methods in the image colour models including hue, saturation and value (HSV) and also hue, saturation and intensity (HSI) models. It should be noted that the first colour model is carried out in the channels of H, I and S entitled HSI-h, HSI-s, and HSI-i, respectively, and also the second colour model is carried out in the channels of H, S and V, entitled HSV-h, HSV-s and HSV-v, respectively.

2 Proposed robust image watermarking framework

The proposed robust image watermarking framework is illustrated in Fig. 1. To discuss the investigated approach, in a better form, at first, the method is schematically presented. There are a number of modules to be processed, while some of them need to be carried out, synchronously. It is to note that the original image as well as the corresponding logo image need to first be acquired. In this regard, the colour model is applied to generate the original image in its HSI or HSV forms, while each one of channels including HSI-h, HSI-s, HSI-i, HSV-h, HSV-s and HSV-v could be processed. The module of contourlet framework that discusses later is realized to present the information, in its new sub-band domain. By calculating the watermarking intensity and also applying the scrambling module that is presented in the form of vector by matrix to vector (M2V) module, the embedding process is carried out. It is needed to note that the binary logo image can be depicted in a sequence of -1 and 1 as

   (1)

Regarding the scrambling module employed in this framework, it should be noted that there are a number of methods to be scrambled the logo images. They are including pseudo random number sequence, Arnold transform and so on. The present Arnold transform is acting in the 2D representation, i.e., A: T²→T².

It is in fact emphasized that the input and the corresponding output dimensionality regarding the present transform is the same as given by

Fig. 1 Schematic diagram of proposed watermarking framework

                      (2)

where N is the log image size and N-1} is assumed. It can be noted that one of the important specifications regarding the present transform is known as its periodic one. Moreover, by applying the inversed contourlet in line with the inversed colour model to the results generated in the output of embedding process, the watermarked image is provided to be analyzed. In order to extract the logo that is now embedded in the cover image in real situations, a number of attacks are organized to be applied to watermarked image. The main core of finding the logo, as an extracted image, is realized in the module of de-embedding process. In general, to obtain the secret data from the cover image, a de-embedding process needs to be organized. In fact, the binary logo image is taken as the outcomes of the aforementioned module. In this framework, the closed loop regarding the attacks is applied to watermarked image, in order to consider the applicability and efficiency of the approach proposed, in an appropriate manner. Hereinafter, the performance of the watermarked image as well as the extracted logo image can be considered through the corresponding module, employed in the proposed framework. Finally,

while the number of attacks is completed, it is the right time to calculate the finalized performance of the system through the investigated finalized performance outcomes that are discussed later. Regarding the simulated attacks, there are a number of intentional and unintentional attacks for images to be realized that are now listed by (1) JPEG compressions in the separated qualities from 10% to 90%, (2) Gaussian noise, impulse noise, salt & pepper noise, (3) median filter in the separated sizes,(4) averaged filter in the separated sizes, crops in the separated sizes, and (5) rotations in the separated angles in the both clock-wise and non-clock-wise directions and finally scaling in the separated values.

2.1 Contourlet module

The proposed contourlet module is schematically illustrated in Fig. 2. Based upon the outcomes, a decision maker system is realized to choose the appropriate levels-directions concerning the contourlet transform to be able to apply to the input information. In the schematic presented, the outcomes are saved and also converting to the tree form to be processed in the proceeding modules.

The idea of realizing the contourlet representation is based on the pyramid directional filter bank (PDFB), which is briefly illustrated in Fig. 3.

Fig. 2 Schematic diagram of contourlet module

Fig. 3 Schematic diagram of pyramid directional filter bank (PDFB)

Figure 3(a) represents the structure of dividing the frequencies with respect to its angles that is used in the output of Fig. 3(b) to describe the main structure of the PDFB that uses the high-low pass filters in association with down sampler module. In order to calculate the pyramid of the image, the input at Jth level is first processed, where the approximation information at (J-1)th level and the corresponding prediction information at Jth level can be acquired. It can be noted that the present mechanism is iterated to reach at 0th level that is now illustrated, schematically, in Fig. 4. Now, with a focus on the contourlet representation, it is organized based on the Laplacian pyramid (LP) and the directional filter bank (DFB) as illustrated in Fig. 5 for an image in the size of N×N. This is initially proposed to represent contours and its textures regarding the image, in an efficient manner.

It is shown that the contourlet is organized by a double filter bank structure to obtain sparse expansions for typical images under smooth contours, while the LP is realized to capture the point discontinuities, at first. Afterwards, a directional filter bank is realized to link the point discontinuities into its linear structures. It should be noted that the overall outcomes regarding the contourlet are presented to be an image expansion via the basic elements such as contour segments. The high frequency (HF) sub-band images are fed from the LP into the DFB to be able to capture the directional information. It is obvious that the contourlet decomposes the image into directional sub-bands at multiple scales, where can be iterated the low frequency (LF) sub-band image, smoothly.

Fig. 4 Schematic diagram of iterated pyramid levels regarding input image

2.1.1 Contourlet embedding mechanism

The embedding mechanism is realized based upon the watermarking intensities as taken as α. This is working to insert the logo information via Eq. (3) by choosing the level and its direction through a decision maker (DM) system in the process of contourlet representation:

                 (3)

where I_L,D(i, j) indicates the sub-bands regarding the contourlet transform in the pyramid level and its direction, since I_WL,D(i, j) indicates the watermarked image, as well. In the approach proposed here, the contourlet coefficients in the form of tree including parent-children are to be modified in a number of levels and the corresponding directions, as presented in Fig. 6.

Fig. 5 Levels and corresponding direction regarding contourlet representation

2.1.2 Levels-directions decomposition

The levels-directions regarding the contourlet based upon the size of the logo image such as 32×32 or 64×64 can be considered to choose, as presented in Fig. 5. The levels can be taken as 2, 3 or 4 and the corresponding direction can also be taken as 1, 2, 3 or 4,respectively. Itis obvious that 24 states in the logo image in the size of 32×32 and 20 states in the size of 64×64 are acquired.

Fig. 6 Contourlet coefficients tree

2.1.3 Decision maker system

The decision maker system is to reach the appropriate decision on the type of the cover image and the corresponding logo image as well as the levels and the corresponding directions decompositions to be watermarked for the purpose of considering the approach performance, in a comparable manner.

2.2 Embedding and corresponding de-embedding modules

The embedding and the corresponding de- embedding modules are schematically illustrated in Figs. 7 and 8, respectively. As mentioned before, the responsibilities of these modules are to embed the logo information into the corresponding cover image to provide the watermarked image and also to extract the logo information from the watermarked image to provide the appropriate watermarking process. To realize the embedding module, it is needed to calculate the max-min information regarding the outcomes of the contourlet module, accurately.

After that, the main core of the embedding process is calculated to be able to update the contourlet results in the chosen levels-directions that are presented through decision maker system.

Fig. 7 Schematic diagram of embedding module

Fig. 8 Schematic diagram of de-embedding module

To concentrate more on the de-embedding module, it is needed to note that the logo image needs to be extracted, at first, and then the results that are presented in the form of vector should be converted to its matrix form through vector to matrix (V2M) module. It is the time to apply the inversed scrambling technique to provide the extracted logo information as illustrated initially. The final stage in the present de-embedding module is to binarize the extracted outcomes as converted in the process of providing the watermarked image.

Regarding the coefficients of the sub-band of the children, which are varied to be embedded are presented by the following, while the value of root node (vorn), the maximum value of child node (mxvocn) and the minimum value of child node (mnvocn) are taken into consideration:

               (4)

where is taken as kth bit of the logo image and α is taken as watermarking intensity. Moreover, regarding the bits of the logo image to be calculated are presented as

                      (5)

2.3 Performance monitoring

2.3.1 Watermarked image analysis

There are a number of performances monitoring, as illustrated, in their schematic forms, in Figs. 9 and 10, respectively, while the first one is related to watermarked image and the second one is related to the corresponding logo image, as well. To realize the first performance monitoring, the mean square errors (MSE), the PSNR and finally the SSIM should be calculated to be presented.

Fig. 9 Schematic diagram of watermarked image performance analysis

Fig. 10 Schematic of logo image performance analysis

The PSNR is presented as

    (6)

where the sizes of the images I_o,w are taken as M×N. Moreover, the SSIM is presented as

     (7)

where δ_x and δ_y are taken as the standard deviations regarding the samples of x and y. Also, δ_xy is taken as the cross correlation and μ_x as well as μ_y are the corresponding averaged values. The parameters C₁, C₂ and C₃ are the constant values.

2.3.2 Logo image analysis

Regarding the second performance monitoring, the normal correlation (NC, C_N) and the bit error rate (BER) should correspondingly be calculated to be presented. It is obvious that the consideration of the outcomes in both cases aims us to verify or to avoid verifying the finalized performance of the framework as analyzed, in its experimental manner, in the proceeding sections in detail.The NC for x and y images is just written by

                 (8)

where the sizes of the images are taken as M×N, as well. Moreover, the BER is traditionally calculated.

3 Experimental results

The applicability and efficiency of the proposed image watermarking framework is here considered through a series of experiments. It is first carried out via a standard cover image, i.e. the pepper and also the logo image that are chosen to be used in these experiments, as illustrated in Fig. 11. It is to note that the present cover image is taken in the size of 512×512, while the logo image is taken in the size of 32×32, as well.

Based upon the results investigated by carrying out the proposed framework in the HSI-i colour model, the outcomes regarding the extracted logo image in the presence of no attack and a number of attacks, reported by the performance monitoring module are tabulated in Table 1, while the watermarking intensity is taken as fifty.

Fig. 11 Standard cover image (a) and logo image (b), used in experiments

Table 1 Results of proposed framework against no attack and a number of simulated attacks in HSI-I model

In order to present the information tabulated in Table 1, Fig. 12 is correspondingly depicted. Please note that the row numbers of Table 1 and Fig. 12 are the same, exactly.

Now, to discuss about the outcomes validity of the framework carried out, the SSIM and the PSNR, reported via performance monitoring module, are tabulated in Table 2 in a number of separated indices for the contourlet embedding representation, while the watermarked intensities are varied in the span of 10-90, respectively.

Fig. 12 Results of proposed framework regarding Table 1:

Table 2 Results regarding watermarking through SSIM and PSNR in three separated images in different levels and directions under a set of intensities values

The outcomes of the performance monitoring with a focus on HSI-h, HSI-i and HSI-s colour models are tabulated in Table 3.

The results in the aforementioned HSI model are considered to extract the logo image against the JPEG compression attacks through the calculation of the NC, as presented in Fig. 13.

The outcomes of the performance monitoring with a focus on HSI-h, HSI-i and HSI-s colour models are tabulated in Table 4.

The results in the aforementioned HSV model are considered to extract the logo image against the JPEG compression attacks through the calculation of the NC, as presented in Fig. 14.

Table 3 Performance monitoring results in HSI-h, HSI-i and HSI-s colour models

Fig. 13 Calculation of NC against JPEG compression attacks in HSI model

Table 4 Performance monitoring results in HSV-h, HSV-s and HSV-v colour models

Fig. 14 Calculation of NC against JPEG compression attacks in HSV model

In order to consider the efficiency of the framework proposed more, the three separated standard cover images including the Lena, the pepper and the baboon are chosen, while both the aforementioned HSI and HSV colour models are concentrated to evaluate via the proposed performance monitoring module. In this way, Fig. 15 illustrates the SSIM for the whole of channels that are synchronously presented to make appropriate sense to be compared.

The PSNR performance monitoring outcomes for the three separated cover images in HSI and HSV colour models are depicted in Fig. 16.

Regarding the logo image analysis, the NC performance monitoring outcomes for the three separated cover images in HSI and HSV colour models are depicted in Fig. 17.

Finally, the BER performance monitoring outcomes for the three separated cover images in HSI and HSV colour models are depicted in Fig. 18.

The results acquired through the proposed watermarking framework in the constant watermarking intensity are now compared with the chosen potential benchmark, as tabulated in Table 5. It should be noted that the idea presented in the aforementioned benchmark is entirely simulated through MATLAB programming language in line with the approach proposed to report the corresponding results.

Table 6 tabulates the comparisons results regarding the proposed framework with respect to the benchmark against a number of the geometrical attacks.

Fig. 15 SSIM performance monitoring outcomes for three separated cover images in HSI and HSV colour models:

Fig. 16 PSNR performance monitoring outcomes for three separated cover images in HSI and HSV colour models:

Fig. 17 NC performance monitoring outcomes for three separated cover images in HSI and HSV colour models:

Fig. 18 BER performance monitoring outcomes for three separated cover images in HSI and HSV colour models:

Table 5 Comparisons of proposed framework with respect to benchmark against a number of attacks

Table 6 Comparisons of proposed framework with respect to benchmark against a number of geometrical attacks

4 Conclusions

A robust image watermarking framework is suggested in the present research to deal with new embedding and de-embedding processes in the area of contourlet transform. The outcomes investigated are to provide the watermarked image and the corresponding extracted logo image, while the high accuracy is taken into real consideration. The suggested framework is of novelty, due to the fact that the integration of modules employed is unique and also the corresponding results are competitive. The idea consists of the contourlet representation, the embedding and the de-embedding modules and finally the performance monitoring. There is also a scrambling module that is working in accordance with levels-directions decomposition in the area of contourlet embedding mechanism, while a decision maker system is organized to deal with the appropriate sub-bands to be embedded in the presence of a series of simulated attacks. Regarding the efficient outcomes, they are reported via the aforementioned performance monitoring that are related to watermarked image as well as the extracted logo image. It should be noted that the first analysis is organized based upon the peak signal-to-noise ratio and the structural similarity indices, as long as the second analysis is also organized based upon the bit error rate and the normal correlation. The present performance monitoring outcomes indicate that the proposed framework is of significance with respect to state-of-the-art.

References

[1] SADREAZAMI H, AHMAD M O, SWAMY M N S. A study of multiplicative watermark detection in the contourlet domain using alpha-stable distributions [J]. IEEE Transactions on Image Processing, 2014, 23(10): 4348-4360.

[2] CHALAMALA S R, KAKKIRALA K R, MALLIKARJUNA R G B. Analysis of wavelet and contourlet transform based image watermarking techniques [C]// IEEE International Conference on Advance Computing (IACC). ITM University Gurgaon, India, 2014: 1122-1126.

[3] IDRISSI N, ROUKHE A. Robust watermarking method based on contourlet transform, maximum entropy, and SVD decomposition [C]// International Conference on Multimedia Computing and Systems (ICMCS). Marrakech, Morocco, 2014: 261-264.

[4] LI Jian-zhong, ZHANG Jun-min. A robust image watermarking scheme with kinoform in hybrid NSCT and SVD domain [C]// 7th IEEE Joint International on Information Technology and Artificial Intelligence Conference (ITAIC). Chongqing, China, 2014: 48-52.

[5] SADREAZAMI H, AHMAD M O, SWAMY M N S. Optimum multiplicative watermark detector in contourlet domain using the normal inverse Gaussian distribution [C]// IEEE International Symposium on Circuits and Systems (ISCAS). Lisbon, Portugal, 2015: 1050-1053.

[6] DUAN Gui-duo, ZHAO Xi, CHEN Ai-guo, LIU Yue-qiang. An improved Hu moment invariants based classification method for watermarking algorithm [C]// International Conference on Information and Network Security, ICINS. Jeju Island, Korea, 2014: 205-209.

[7] SADREAZAMI H, AHMAD M O, SWAMY M N S. Contourlet domain image modeling by using the alpha-stable family of distributions [C]// IEEE International Symposium on Circuits and Systems (ISCAS). Melbourne VIC, Australia, 2014: 1288-1291.

[8] DUAN Gui-duo, LIU Yue-qiang, ZHAO Xi. An improved watermarking algorithm using adaptive threshold [C]// International Conference on Progress in Informatics and Computing (PIC). Shanghai, China, 2014: 385-389.

[9] RANJBAR S, ZARGARI F, GHANBARI M. A highly robust two-stage Contourlet-based digital image watermarking method [J]. Signal Processing: Image Communication, 2013, 28(10): 1526-1536.

[10] NIU Pan-pan, WANG Xiang-yang, YANG Yi-ping, LU Ming-yu. A novel color image watermarking scheme in nonsampled contourlet-domain [J]. Expert Systems with Applications, 2011, 38(3): 2081-2098.

[11] WANG Xiang-yang, WANG Ai-long, YANG Hong-ying, ZHANG Yan, WANG Chun-peng. A new robust digital watermarking based on exponent moments invariants in nonsubsampled contourlet transform domain [J]. Computers & Electrical Engineering, 2014, 40(3): 942-955.

[12] SONG Hao-hao, YU Song-yu, YANG Xiao-kang, SONG Li, WANG Chen. Contourlet-based image adaptive watermarking [J]. Signal Processing: Image Communication, 2008, 23(3): 162-178.

[13] SONG Wei, HOU Jian-jun. Chaotic system and QR factorization based robust digital image watermarking algorithm [J]. Journal of Central South University of Technology, 2011, 18(1): 116-124.

[14] RAHIMI F, RABBANI H. A dual adaptive watermarking scheme in contourlet domain for DICOM images [J]. Bio Medical Engineering, 2011, 10(53): 1-18.

[15] WANG Tai-yue, LI Hong-wei. A novel scrambling digital image watermarking algorithm based on contourlet transform [J]. Wuhan University Journal of Natural Sciences, 2014, 19(4): 315-322.

(Edited by YANG Bing)

Cite this article as: M. F. Kazemi, M. A. Pourmina, A. H. Mazinan. A new image watermarking framework based on levels-directions decomposition in contourlet representation [J]. Journal of Central South University, 2017, 24(3): 521-532. DOI: 10.1007/s11771-017-3455-3.

Received date: 2015-05-19; Accepted date: 2015-10-08

Corresponding author: A. H. Mazinan, Associate Professor; Tel: +98-21-8883082630; Fax: +98-21-88830831; E-mail: ahmazinan@gmail.com, ah_mazinan@yahoo.com, mazinan@azad.ac.ir