School of Information Science and Technology,Northwest University,Xi’an 710069,China

Digital watermarking algorithm based on scale-invariant feature regions in non-subsampled contourlet transform domain

Jian Zhao*,Na Zhang,Jian Jia,and Huanwei Wang

School of Information Science and Technology,Northwest University,Xi’an 710069,China

Contraposing the need of the robust digital watermark for the copyright protection field a new digital watermarking algorithm in the non-subsampled contourlet transform(NSCT)domain is proposed.The largest energy sub-band after NSCT is selected to embed watermark.The watermark is embedded into scaleinvariant feature transform(SIFT)regions.During embedding,the initial region is divided into some cirque sub-regions with the same area,and each watermark bit is embedded into one sub-region. Extensive simulation results and comparisons show that the algorithm gets a good trade-off of invisibility,robustness and capacity, thus obtaining good quality of the image while being able to effectively resist common image processing,and geometric and combo attacks,and normalized similarity is almost all reached.

multi-scale geometric analysis (MGA), nonsubsampled contourlet transform(NSCT),scale-invariant feature region.

1.Introduction

The past few years have seen an explosion in the use of the digital watermarking technology in copyright protection[1].And digital watermarking for this purpose must have robustness,security and so on[2].Thus,the digital watermarking algorithm based on multi-scale geometric analysis(MGA),which is more effective than wavelet analysis to capture the image edge and texture details[3], is getting developed in recent years.Rama et al.presented a watermarking algorithm based on ridgelet transform[4]. Kaviani et al.embedded digital watermark in the contourlet domain[5].These algorithms taking advantages of MGA[6–10]are better than many digital watermarking algorithmsbased on discrete wavelet transform(DWT) [11–14].However,the watermarking algorithm based on contourlet transform[15,16]has small redundancy,which is not conduciveto selecting the watermarkembeddingposition and has no translational invariance[17].To overcome these shortcomings,the watermarking algorithm with rotation,scaling,translation invariance and high robustness and invisibility is urgent to be proposed.

Facing the above problems,a digital watermarking algorithm based on the non-subsampled contourlet domain is provided in this paper on the theory of MGA[18].

2.Theories of NSCT

Akhaee et al.proposed a new non-adaptive multi-scale geometric representation of image-contourlettransform[19], which is a multi-resolution,multi-direction and local image representation.But it has small redundancy,and it is not conducive to compressing and selecting the watermark embedding position;the upper and lower samplings exist in Laplace pyramid(LP)and the directional filte banks(DFB)make the contourlet transform do not have the translational invariance.To overcomethese shortcomings,Cunha proposed a translation invariance,multiscale,multi-direction conversion non-subsampled contourlettransform(NSCT)[20].Thereisnoupperandlower sampling,so all sub-bands are the same size as the original image,while its redundancy provides us more coeffi cients to embed watermarks.Their structure and frequency schematic diagram are shown in Fig.1.Fig.2 shows the non-sampling filte tower and three-dimensional exploded view schematic sub-band frequency.

Fig.1 Transformation diagram of NSCT

Fig.2 Three-stage decomposition of NSCT and frequency plane

3.Proposed watermark embedding and extraction algorithm

Because of NSCT with multi-scale,multi-direction and translation invariance,embedding watermark can embed watermark in more detail information.The scale-invariant feature transform(SIFT)is scale invariant,rotation invariant,and local descriptors[21]to achievea goodwatermark locality.So this paper proposes a novel digital watermarking algorithm based on SIFT in NSCT.Extracting SIFT feature points to construct feature regions,watermark embeddedness and extraction are conducted in the NSCT domain.

3.1Proposed watermarking algorithm

The following details are about the odd-even quantization algorithm embedding watermark.

Quantization function is used to assign eachI(x,y)to symbol“0”or“1”:

whereΔdenotes the quantization step.

Firstly,calculate the quantization noise of quantization regions:

Then,the coefficien is determinedby the following formula:

whereωirepresents the watermarking value.Finally,the modifie NSCT coefficien value ofI?(x,y)is

3.2Procedure of the proposed embedding and extraction scheme

Before the watermark embedded,the watermark image is preprocessed[22].

(i)Extract SIFT feature points of the carrier image and constitut stable circular regions;

(ii)The characteristic regions conduct two layers of NSCT;

(iii)Calculate energy of the second layer in all directions after NSCT decomposition,and choose the largest sub-band energy pattern to embed watermark and record the sub-band chart,the formula that calculating the energy is as follows:

(iv)The selection of the directional sub-band is divided into cirque sub-region to embed watermark.The formulas used in the process of constituting the concentric annular characteristic region are(6)and(7).Each ring embeddeda watermark signal using the algorithm proposed in Section 3.1,is shown in Fig.3.

(v)The above steps are done circularly for all feature regions,producing the whole watermarked image.

Fig.3 Cirque sub-region to embed watermark

(vi)The embeddedwatermarkpattern instead of the original sub-band pattern,can restore the carrier image after inverse NSCT transform,completing watermark embedding.

In this paper,the watermark extraction algorithm is the inverse process of the watermark embedding process.

The quantization function used in the embedding process is used to calculate the quantization function value of the coefficient within the region,and note that the number of the function value equalled 0 isNUM0the number of the function value equalled 1 isNUM1,by the following formula to extract the watermark:

4.Experimental results and discussion

Many images are used in experiments in this paper.Here taking“Lena”and“peppers”gray images as examples. The original image is the size of 512×512,the watermark image is a size of 32×32 binary image.The feature region for a 2-level NSCT transform,odd-even quantization step isΔ=6.Fig.4 showsthe watermarkimages,Fig.5shows the original images.

Fig.4 Watermark images

Fig.5 Original images

Fig.6 shows the sub-bands of 2-level NSCT of the original image(here taking“Lena”as an example).

Fig.6 Sub-bands of 2-level NSCT of“Lena”

4.1 Invisible watermark

Fig.7 describes the contrast of the original image of“Lena”and the watermarked image;Fig.8 illustrates the contrast of the original image of“peppers”and the watermarked image.Fig.7 and Fig.8 show that there is no significan difference between the original image and the watermarked image from the subjective perspective.Meanwhile,Fig.7 and Fig.8 use“Lena”and“peppers”images respectively,it can be seen that different host images have good results,showing that the proposed algorithm applied to the host image have different textures.

Fig.7 Contrast of“Lena”

Fig.8 Contrast of“peppers”

With thepeaksignalto noiseratio(PSNR)[23]toevaluate invisibility from the objective aspect,PSNR represents an objective evaluation of the degree of the image’s deterioration.The PSNR value between the two images is larger which tends to be no deterioration.It is define as

whereI′(i,j)represents the watermarked image,I(i,j) representstheoriginalimage,andtheimagesizeisM×N.

In this paper,a number of experiments like embedding watermark and attacking the watermarked image are performedwith typical digital watermarkingalgorithmsbased on the DWT domain[24–26]or contourlet transform domain[27–29].Meanwhile,the results of experiments above are compared with the result of the proposed algorithm based on NSCT.Literature[25]and literature[27] regarded as examples are shown here.This algorithm’s PSNR(dB)is shown in Table 1.This table also lists the literature[25]andliterature[27]algorithms’PSNR values.

Table 1 Values of watermarked PSNR dB

Table 1 shows that PSNR values obtained in the proposed algorithm is much larger than those in the literature [25]and[27]algorithms’,the proposedalgorithm has very good invisibility.Because the proposed algorithm based on the NSCT domain,which has multi-scale and multidirection,can be a good description of the outline and detail information of an image.When embedded watermark information,watermark can be hid in the carrier image for more detailed information,thereby enhance the invisibility.The proposed algorithm and literature algorithms’fig ure“Lena”obtained PSNR values larger than the figur“peppers”,because the image of“Lena”has more texture and detail information than the image of“peppers”.The advantagesof the proposedalgorithmin the embeddedwatermark are also illustrated.

4.2Robust watermarking

Available normalized similarity(NC)[30]is selected to evaluate the robustness.ω′(i)represents the extracted watermarkimage,andω(i)means the original watermarkimage.The larger the NC value between the two images,the better the robustness.It is define as

Fig.9 showsthe correspondingextractedwatermarkimage after watermarked the image subjected to conventional signal processing,geometric attack and combinations attack.

Fig.9 Corresponding extracted watermark image after watermarked the image subjected to different attacks

The extracted watermark images are all successfully detected.From a visual point of view,the algorithm has good robustness.Because in this paper watermark embedding and extraction are achieved in the NSCT domain, which uses lines mentioned in multi-scale geometric analysis techniques instead of separate points to capture contours,thus improve the watermark extraction quality and get the rapid recovery watermark contours when the watermarked image is attacked.

Table 2 shows the values of the NC of the extracted watermarkwhenthe watermarkedimage subjectedto conventional signal processing and geometric attacks,with algorithms of this paper,the literature[25]and literature[27].

Table 2 reflect that when there is no attack,the NC is 1. The watermark extraction algorithm has good reversibility and can accurately extract the watermark.Vertically to see suffering various attacks,NC values are all greater than 0.9.Transverse to see whether“Lena”or“peppers”hasa very high NC value.It can be seen from the NC value comparison that,the proposed algorithm’s is larger than the literature algorithms’,so that the proposed algorithm has a very good robustness.It can be seen from Table 2 that when the center cuts 10%,NC value reaches 1.This is because it has locality in the NSCT domaincombinedwith the SIFT feature region;NC value is also 1 when suffering rotation.The SIFT region has rotational invariance,so do cirque feature regions.Further to see,it can resist scaling attack because of scale invariance in the embedding region,it can be also completely extracted when suffering the zoom.

The results of the proposed algorithm,and literature algorithms are compared.It can be seen from Table 1 and Table 2 that the literature[25]’s PSNR is higher than the literature[27]’s,but the literature[25]’s NC is lower than the literature[27]’s.However,the PSNR and NC of the proposed algorithm are both higher than the literature’s. The comparative literature experiments and results prove that the proposed algorithm gets a good trade-off of invisibility and robustness.

Table 2 Values of NC of the watermarked image suffered attacks

5.Conclusions

Thispaperpresentsadigitalwatermarkingalgorithmbased on NSCT.The core of this algorithm lies at embedding watermark in the NSCT domain,and combining with the SIFT cirque feature region.Due to its multi-scale,multidirection and translation invariant characteristics,it can embed watermark in more detail information areas which are not sensitive to the vision,thus getting good invisibility.And it can quickly restore the contour lines,which improvesthe quality of extracted watermark.The SIFT descriptor are scale,rotation invariant,local descriptors,so it can achieve good watermark locality.Experimental results show that the proposed method can not only obtain good qualityofthe image,but also effectivelyresist attacks from commonsignalprocessingandgeometricattacksincluding rotation,zoom,and translation.

[1]J.Zhao.Information technology:watermarking by numbers.Nature,1996,384(10):514–515.

[2]A.Darwish,A.Abraham.The use of computational intelligence in digital watermarking.Review,Challenges,and New Trends,2011,21(4):277–297.

[3]E.I.Sami,L.Z.Baba.Watermarking of digital images in frequency domain.International Journal of Automation and Computing,2010,7(1):17–22.

[4]S.R.C.Rama,V.N.K.Prasad.Digital watermarking based on magic square and ridgelet transform techniques.Intelligent Systems and Computing,2014,243(24):143–161.

[5]H.R.Kaviani,N.Karimi.Robust watermarking in singular values of contourlet coefficientIranian Conference on 7th Machine Vision and Image Processing,2011:13–15.

[6]L.D.Li,X.P.Yuan,L.L.Zhao.Rotation invariant watermark embedding based on scale-adapted characteristic regions.Information Science,2010,180(2):2875–2888.

[7]X.Y.Wang,A.L.Wang,H.Y.Yang.A new robust digital watermarking based on exponent moments invariants in nonsubsampled contourlet transform domain.Computers&Electrical Engineering,2014,40(3):942–955.

[8]V.Seenivasagam,R.Velumani.Inversion attack resilient zerowatermarking scheme for medical image authentication.Image Processing,2014,8(12):718–727.

[9]L.D.Li,B.L.Guo,J.F.Biao.Spatial domain image watermarking scheme robusttogeometricattacks based onodd-even quantization.Journal of Electronics&Information Technology,2009,31(2):134–138.

[10]C.Deng,J.Li,X.B.Gao.The watermarking algorithm based on the anti-affin covariant regions geometric attacks.Institute of Automation,2012,36(2):221–228.

[11]R.A.L.Mohammad,G.D.M.Sedigheh.Using curvelet transform for watermarking based on amplitude modulation signal.Image and Video Processing,2014,8(4):687–697.

[12]C.Ye,Y.Q.Shen,H.Li.A wavelet transform(DWT)digital watermarking algorithm based on human visual system(HVS) discrete.Journal of Zhejiang University,2013,40(2):153–156.

[13]X.Zou,K.Li.Digital image watermarking algorithm resistant togeometric attacksinwaveletdomain.Journal ofInformation&Computational Science,2012,12(9):3391–3399.

[14]L.Li,X.Yuan,Z.Lu.Rotation invariant watermark embedding based on scale-adapted characteristic regions.Information Sciences,2010,180(15):2875–2888.

[15]S.Hamidreza,M.A.Omair,M.N.Swamy.A study of multiplicative watermark detection in the contourlet domain usingalpha-stable distributions.Image Processing,2014,23(10): 4348–4360.

[16]S.M.Zhu,J.M.Liu.Adaptive quantization index modulationdigitalwatermarking algorithmbased on conrtourlet transform.Acta Optica Sinica,2009,29(6):1523–1529.

[17]R.Saei,Z.Farzad,G.Mohammed.A highly robust two-stage Contourlet-based digital image watermark ingmethod.Signal Processing:Image Communication,2013,28(10):1526–1536.

[18]L.L.Jiao,S.Wang,F.Liu.Image multi-scale geometric analysis theory and application.Xi’an:Xi’an University of Electronic Science and Technology Press,2008.

[19]M.A.Akhaee,S.M.E.Sahraeian,F.Marvasti.Contourletbased image watermarking using optimum detector in a noisy environment.Image Processing,2010,19(4):967–980.

[20]A.L.Cunha,J.Zhou.The non-subsampled contourlet transform:theory,design,and applications.IEEE Trans.on Image Process,2006,5(12):3089–3101.

[21]D.G.Lowe.Distinctive image features from scale-invariant key points.International Journal of Computer Visio,2004: 271–279.

[22]W.W.Zhang,F.Gao,B.Liu.A watermark strategy for quantum images based on quantum Fourier transform.Quantum Information Processing,2013,12(2):793–803.

[23]X.Y.Wang,A.L.Wang,H.Y.Yang.A new robust digital watermarking based on exponent moments invariants in nonsubsampled contourlet transform domain.Computers&Electrical Engineering,2014,40(3):942–955.

[24]P.Lu,X.Y.Liu,X.Lu.A digital watermark algorithm based on digital wavelet transform and human visual system.Information Technology,2008,32(10):35–37.

[25]L.L.Chen,X.Y.Wang.Arobust digitalwatermarking scheme based on Elliptical area of SIFT.Computer Engineering and Applications,2012,48(4):98–101.

[26]B.Zhang,J.Z.Wang,J.F.Wen.The watermarking algorithm reaearch and implementationbased on scale-invariant feature transform against geometric attacks.Computer Technology and Development,2011,21(3):175–178.

[27]N.C.H.J.Ma.The robust watermarking algorithm based on contourlet transform and singular value decomposition.Application Research of Computers,2012,29(7):2700–2702.

[28]N.Li,E.Hancock,X.S.Zheng,et al.Improved contentbased watermarking using scale-invariant feature points.Lecture Notes in Computer Science,2011,6978:636–649.

[29]C.W.Zhang,W.X.Zhang.The contour matching watermarking algorithm based on contourlet transform.Computer Applications and Software,2012,29(11):119–123.

[30]H.Berghel,L.O.Gorman.Protecting ownership rights through digital watermarking.Computer,1996,29(7):101–103.

Biographies

Jian Zhaowasborn in1973.Hereceived hisPh.D. degree in signal and information processing from Northwestern Polytechnical University.And then from 2004 to 2006,he completed his postdoctoral thesis in Computer Science and Technology.From 2006 to 2008,he completed his postdoctoral thesis in Electronic Science and Technology.He is also a professor in Northwest University.His research interests include signal and information processing,computer science, internet of things,and information security.

E-mail:zjctec@nwu.edu.cn

Na Zhangwas born in 1989.She received her B.S. degree from Yan’an University.She is currently pursuing her M.S.degree in communication and information system from Northwest University.Her research interests include image processing,digital watermarking technology and applications.

E-mail:852589910@qq.com

Jian Jiawas born in 1977.He received his M.S.degree in mathematics from Northwest University in 2001,and his Ph.D.degree in pattern recognition and intelligent systems from the Institute of Intelligent Information Processing of Xidian University in 2008.He is currently an associate professor in Northwest University.His current research interests include image processing and wavelet analysis.

E-mail:jiajian@nwu.edu.cn

Huanwei Wangwas born in 1987.He received his B.S.degree from Northwest University.He is currently pursuing his M.S.degree in information and signal processing from Northwest University.His main research interests include image processing, machine vision,and two-dimensional code defect detection.

E-mail:wanghw1003@126.com

10.1109/JSEE.2015.00143

Manuscript received May 06,2014.

*Corresponding author.

This work was supported by the National Natural Science Foundation of China(61379010)and the Natural Science Basic Research Plan in Shaanxi Province of China(2015JM6293).