An Efficient Image Watermarking Method Based on Fast ...downloads.hindawi.com/journals/mpe/2017/3509258.pdfbut the encryption is time consuming. The above SVD-basedwatermarkingmayhavegoodrobustnessagainstJPEG

Research ArticleAn Efficient Image Watermarking Method Based onFast Discrete Cosine Transform Algorithm

S E Tsai1 K C Liu2 and S M Yang3

1Department of Computer Science and Information Engineering Chang Jung Christian University Tainan City 701 Taiwan2Asian Institute of Telesurgery Changhua County 500 Taiwan3Department of Aeronautics and Astronautics National Cheng Kung University Tainan City 701 Taiwan

Correspondence should be addressed to S E Tsai seanmailcjcuedutw

Received 15 February 2017 Revised 24 May 2017 Accepted 21 June 2017 Published 16 August 2017

Academic Editor Raffaele Solimene

Copyright copy 2017 S E Tsai et alThis is an open access article distributed under the Creative Commons Attribution License whichpermits unrestricted use distribution and reproduction in any medium provided the original work is properly cited

This paper proposes an image watermarking method based on the fast discrete cosine transform (DCT) algorithm forimplementation in digital signal processor A digital watermark can be effectively embedded and efficiently extracted without thehost image The keys in watermarking process include four frequency coefficients in DCT two random permutation vectors anda quantization matrix for normalizing the watermark and the host image The fast DCT algorithm has been shown to reduce thecomplexity of two-dimensional image transformation so that embeddingdecoding an image watermark can be completed in realtimewithin 033 secondsThe quality of bothwatermarked image and extracted (retrieved) watermark remains excellent It is shownthat the watermarking method is efficient in and robust to data cropping transmission loss and compressiondecompression

1 Introduction

Conventional technique of embedding watermark signals orpatterns into digital image is to insert a secret bit stringin spatial frequency or wavelet domain In spatial domainwatermarking the human visual system (HVS) is applied toassure that modifications of an original image by embeddedwatermark are imperceptible Koz and Alatan [1] appliedthe temporal contrast threshold of HVS to prevent imagedistortion after watermarking but the watermarked imageis vulnerable to data transmission Kaur et al [2] proposeda technique to substitute the blue component of imagepixels with character bits but it is also vulnerable to signalprocessing attack Agarwal [3] embedded a watermark withdata encryptionwhich required substantial data capacity thusleading to low image quality Tyagi et al [4] also appliedthe least significant bit (LSB) to achieve high security atthe price of lower image quality Nath et al [5] proposed amethod for hiding encrypted secret message inside a coverfile which again is vulnerable to data compression and attackSara et al [6] selected the pixels with higher intensity to

embed watermark bits by LSB A major constraint of theabove studies is that they all require the original image datato decode the watermark

In ldquoblindrdquo watermarking Chadha et al [7] proposed amethod based on LSB and discrete cosine transform (DCT)for decoding watermark without the original image butthe method is sensitive to common image compressionSurrah and Mohamed [8] presented a binary data hidingalgorithm by changing the position of every image blockto increase security Wang et al [9] proposed watermarkingby quaternion Fourier transform but the image quality wasreduced significantly by the embeddedwatermarkWith JointPhotographic Experts Group (JPEG) adopting DCT to itscompression process many watermarking methods based onDCT have been developed Benoraira et al [10] proposedembedding the watermark bits sequence in both DCT anddiscrete wavelet transform (DWT) to guard against attackbut the procedure is computational intensive A watermark-ing technique based on the so-called wavelet-tree [11] andanother based on DWT [12] were also proposed achieving

HindawiMathematical Problems in EngineeringVolume 2017 Article ID 3509258 10 pageshttpsdoiorg10115520173509258

2 Mathematical Problems in Engineering

better robustness however both required substantial compu-tation with large data capacity

Singular value decomposition (SVD) is desirable forimplementation together with DCT or DWT because itrequires extensive computations if applied separately ontoimages Rawat and Raman [13] proposed an SVD-basedwatermarking scheme to increase security in copyright pro-tection Makbol and Khoo [14] and Lagzian et al [15] alsodeveloped image watermarking schemes based on DWT andSVD to achieve robustness against image processing but thecomputational load was substantial Hu and Hsu [16] pro-posed image watermarking with DWT-SVD-DCT featuresto guard against compression Mishra et al [17] applied afirefly algorithm based on DWT and SVD for watermarkingbut the encryption is time consuming The above SVD-based watermarking may have good robustness against JPEGcompression however the embedded watermark becomesvulnerable because all pixels would be changed when onesingular value is modified This paper presents an efficientdigital image watermarking method based on the fast DCTalgorithm for real-time application It is shown that thewatermarked image quality robustness and computationloading can all be improved

2 Embedding Watermark by the FastDCT Algorithm

Consider an 8-bit gray-level original image S and an imagewatermarkW

S = 119904 (119894 119895) 0 ≦ 119894 119895 ≦ 119873 minus 1 0 ≦ 119904 (119894 119895) ≦ 255 W

= 119908 (119901 119902) 0 ≦ 119901 119902 ≦ 119872 minus 1 0 ≦ 119908 (119901 119902) ≦ 255 (1)

where119872 and119873 are integers119872 lt 119873 The original image S isdivided into nonoverlapping pixel blocks of size 8times8 and thentransformed into frequency domain Each block is processedindependently by two-dimensional (2D) DCT into matrix FThe fast DCT algorithm [18] is applied to decompose the 2DDCT into a pair of 1D DCTs for computation efficiency Fora digital image S the 2D DCTmatrix F in frequency domainand its inverse transform back to spatial domain S1015840 can bewritten in matrix form by

F = M S M119879 (2a)

S1015840 = M119879 F M (2b)

where matrixM = 119898(119906 V) with 119898(119906 V) being the elementsin the 119906th row and Vth column

119898(119906 V)

=

12radic2 119906 = 0 cup 0 le V le 712 cos (2V + 1) 11990612058716 1 le 119906 le 7 cup 0 le V le 7

(3)

The 2D spatial datamatrix S1015840 in (2b) is the linear combinationof the base images which are obtained by the outer productof the column and row vectors of M By exploiting theredundancy in DCT coefficients it can be shown that the fastDCT algorithm can reduce the complexity of a 2D DCT of 8times 8 block from S to F in (2a) in only 24 multiplications andthe inverse transform from F to spatial domain S1015840 in (2b) issimply the transpose of the DCT matrixM [18]

Watermarking is performed first by decomposing a truecolor image into RGB components and the watermarkshould be embedded in the main composition of the orig-inal image to maintain image quality after watermarkingFor example an image rich in blue shall be watermarkedwith gray-level blue watermarks Due to the lower sen-sitivity of human eyes to blue light it is recommendedto embed watermark in blue component if one cannotdetermine the main composition of the image The imageembedding process of the watermarking method as shownin Figure 1 includes (1) computation optimization by thefast DCT algorithm and its inverse transform (2) selectionof the middle frequency coefficients (3) permutation ofthe watermark and (4) quantization (normalization) of thewatermark and the host image A spatial domain image S isprocessed by the fast DCT algorithm into spectral domainF Intuitively a watermark should be embedded into thehigher frequency coefficients in DCT for better perceptualinvisibility However the image ldquoenergyrdquo is concentrated inthe low frequency coefficients (eg the 0thsim5th coefficientsin Figure 2) so that the information embedded (hidden) inthe high frequency coefficients (the 28thsim63th coefficients)may be vulnerable to data processing Tomaintain the qualityof a host image after watermarking and to improve therobustness of a watermarked image against data processingit is recommended to embed the watermark in some of themiddle frequency coefficients (the 6thsim27th coefficients) asshown in the light color region of Figure 2 Because the hostimage of size 119873 times 119873 and the watermark119872 times119872 are dividedinto nonoverlapping pixel blocks of size 8 times 8 the number ofthe middle frequency coefficients selected for embedding thewatermark is (8119872119873)2 such that the sizes of the host imageand thewatermark are compatible For an 8-bit original imageof size 256times 256 (119873 = 256) and 8-bit digital watermark of size64 times 64 (119872 = 64) 4 middle frequency coefficients in each ofthe 8 times 8 blocks from the host image are needed to embedthe watermark The image block composed of these middlefrequency coefficients is denoted as Fm in frequency domain

A random number generator is applied to simultaneouslygenerate two vectors to permute the watermark W into Wpin spatial domain These vectors with elements in [1119872]representing the new coordinates (119909 119910) serve as a key towatermark decoding The permuted watermark is dividedinto nonoverlapping 4 times 4 blocks and processed by the fastDCT algorithm into Wf in spectral domain Quantization isrequired to normalize the image intensity of the host imageandwatermark thereby increasing the quality ofwatermarkedimage and reducing the watermarkmemory sizeThe processis to scale the watermark Wf into Wq by the host imageproperty represented by the average of the middle frequency

Mathematical Problems in Engineering 3

True color image

R B

Recombination of theRGB components

Watermarked color image

G

Choose one of the components (R G or B) at will

Originalimage

(1) The fast DCT algorithm

Watermark (3) Random

(2) Selection of the four middlefrequency coefficients

Modification of the middle frequency coefficients by the watermark

Watermarked image

(4) QuantizationFast DCT algorithm

The fast IDCTalgorithm

S F

W

Decomposition

FmFw

Fw

WpWf Wq

mdash

permutation

Sw

Figure 1 The watermarking embedding process that integrates (1) the fast DCT algorithm (2) selection of the four middle frequencycoefficients (3) random permutation of the watermark and (4) normalization by quantization


0 1 5 6 14 15 27 28

2 4 7 13 16 26 29 42

3 8 12 17 25 30 41 43

9 11 18 24 31 40 44 53

10 19 23 32 39 45 52 54

20 22 33 38 46 51 55 60

21 34 37 47 50 56 59 61

35 36 48 49 57 58 62 63

Figure 2 Distribution of coefficients in DCT (a) low frequency(heavy color) (b) middle frequency (light color) and (c) highfrequency (white)

coefficients denoted as 119865ave The element of quantizationmatrixQ = 119876(119886 119887) is defined119876 (119886 119887) = 161198722119865ave (

1198724minus1sum119894=0

1198724minus1sum119895=0

Wf (119886 + 4119894 119887 + 4119895)) (4)

such that Wf is scaled element by element or block by blockinto Wq Note that the watermark Wq now contains thehost image information 119865ave so the impact on image qualityafter watermarking will be minimized When embedding thewatermark image the four coefficients selected in Fm will besubstituted by the corresponding element inWq into Fw Thefinal step is inverse transformofFw by the fastDCT algorithmin (2b) into watermarked image S1015840w in spectral domain Insummary the parameters in the embedding process are thefour middle frequency coefficients two random vectors andone quantization matrix for normalization All are to bestored as keys for later decoding process

A quantitative measurement for evaluating the fidelity ofthe watermarked image S1015840w is the peak to signal noise ratio

PSNR = 10 log10 (2552MSE) (5)

where MSE is the mean-square error For an image of blocksize 8 times 8 its MSE is

MSE = 1647sum119894=0

7sum119895=0

(119904 (119894 119895) minus 1199041015840119908 (119894 119895))2 (6)

where 119904(119894 119895) and 1199041015840119908(119894 119895) represent the pixel value in theoriginal image S and watermarked image S1015840w respectivelyThe higher the PSNR the better the image quality In generalhuman visual system cannot differentiate a watermarkedimage from the host image when PSNR ge 30 dB3 Decoding Watermark

The step of decoding process as shown in Figure 3 is reverseto the embedding process A watermarked image is first

divided into nonoverlapped blocks of size 8 times 8 and eachblock is processed by the fast DCT algorithm The middlefrequency coefficients in each block are extracted by usingthe key in the embedding process Inverse quantization ofthe extracted watermark is by matrix Q = 119876(119886 119887) from thekey in embedding process Inverse DCT can then be appliedto obtain the watermark image in spatial domain Reversepermutation by using the key of the two random vectors isto extract the watermark embedded in the image denotedas W1015840 Note that no original image is required in decodinga watermarked image

The normalized correlation (NC) is defined to evaluatethe fidelity of the extracted watermark

NC = sum119872minus1119894=0 sum119872minus1119895=0 119908 (119894 119895) 1199081015840 (119894 119895)sum119872minus1119894=0 sum119872minus1119895=0 (119908 (119894 119895))2 (7)

This is to represent the cross-correlation of the referenceand extracted watermarks If the extracted watermark1199081015840(119894 119895)is the same as the reference (registered) watermark image119908(119894 119895) then NC = 10 Note that NC is a quantitativemeasurement it is not necessarily related to image qualityNC = 10 indicates that the watermarking method can extractthe embedded watermark faithfully In summary the stepsof decoding process are (1) selection of the RGB componentof the watermarked image (2) computation by the fast DCTalgorithm (3) extraction of themiddle frequency coefficientsand (4) inverse quantization inverse DCT and reversepermutation

4 Watermarking Implementation

41 PC-Based Watermarking In order to validate the pro-posed watermarking method experiments are conducted oncomputer (CPU AMDAM3 Athlon II X4-640 RAM DDR34G-1333 and VGA MSI R5770 HAWK 1G GDDR5) in Win-dows 8 with Matlab 2012 The digital image can be storedin Windows Bitmap (BMP) Graphics Interchange Format(GIF) Joint Photographic Experts Group (JPEG) or TaggedImage File Format (TIFF) The use of color image processingis motivated by two factors (1) color is a powerful descriptorthat often simplifies object identification and extraction froma scene and (2) human can distinguish thousands of colorshades and intensities compared to about only two dozenshades of gray From the absorption characteristics of humaneyes colors are seen as variable combinations of the primarycolors red (R) green (G) and blue (B) In practice a colormodel is a specification of coordinate system within whichthe color is represented by a single point

By the image watermarking method illustrated in Fig-ure 1 the three components of an image are stored indata array that defines RGB components for each individualpixel Embedding a watermark is dependent on the maincomposition of the color image An original image of size 256times 256 and a digital watermark of 8-bit gray-level of size 64 times64 as shown in Table 2 are to demonstrate the watermarkingmethod The original image divided into nonoverlap blocksof size 8 times 8 shall select 4 coefficients in the middle frequencyrange after DCTThemiddle range coefficients in coordinates


R

B

Watermarkedcolor image

G

Selection of one component

The Fast DCTalgorithm

Registeredwatermark

Reverse permutation

Inverse quantization andthe fast IDCT algorithm

Extraction of the middlefrequency coefficients

Extractedwatermark

Normalizedcorrelation

W

Decomposition

W

Figure 3 Watermark decoding process by the fast DCT algorithm with the key of (1) the four middle frequency coefficients (2) the randompermutation vectors and (3) the quantization matrix

(1 4) (4 2) (3 3) and (1 5) corresponding to block numbers6 11 12 and 14 are selected to embed the image watermarkTable 1 also lists the two random permutation vectors andthe quantization matrix 119876 of (4) The experiment result isshown in Table 2 where PSNR = 31256 dB and NC = 10and the watermark embeddingdecoding can be completedwithin 16 seconds The result indicates that the proposedwatermarking method can maintain the image quality andthe embeddingdecoding process is efficient

Experiments are also conducted to validate the robustnessof the watermarking method against image cropping anddata compression Table 3 shows that after 5 cropping ofwatermarked image NC = 0931 For 15 and 25 croppingthe watermarked image yields NC = 0777 and 0707 respec-tively and the quality of the retrieved watermark remainssatisfactory These results show that selection of the fourmiddle frequency coefficients for embedding watermark is

robust to data compression Although some distinct featuresin the retrieved watermark are lost after 25 cropping or 14data compression the method can still correctly extract andidentify the watermark

Experiments of color image embedding in different com-ponents are also conducted For the color image in Table 4the watermark can be embedded into R G or B componentin themiddle frequency coefficientsThe watermarked in anyof the components can be completed within 2 seconds withPSNR higher than 41 dB

42 DSP-Based Watermarking With rapid development ofdigital signal processor (DSP) implementation of real-time processing is feasible The watermarking method ismost effective when implemented in digital signal processor(TMS320C6701) which includes eight functional units CPUinstruction packing to reduce code size 40-bit arithmetic


Table 1 The key in watermark embeddingdecoding process includes (1) four frequency coefficients (2) two random permutation vectorsand (3) a quantization matrix

(1) Frequency coefficients 6 11 12 14

(2) Random permutation vectors

[44 19 34 9 62 7 50 57 39 20 40 30 10 49 53 27 52 43 6 13 4 26 23 47 55 3638 64 58 61 21 14 45 41 8 59 12 18 24 31 1 32 11 29 56 48 42 15 5 46 63 51

25 37 28 2 54 21 60 3 16 33 35 17][57 47 39 40 3 32 15 50 37 64 56 36 23 11 63 42 9 51 46 10 2 30 61 60 55 4143 59 26 25 27 4 35 45 20 53 7 12 8 31 21 44 62 5 33 16 54 6 22 18 14 17 13

1 19 49 29 52 58 38 34 28 48 24]

(3) Quantization matrix Q

[[[[[[[[

9882351 minus64521 11211 minus1112963722 36798 minus06523 minus30758minus63219 16128 minus09834 minus1724534911 11675 minus15218 23147

]]]]]]]]

Table 2 Validation of the watermarking method where the watermarked image remains in excellent quality with PSNR ge 30 dB and theretrieved watermark with perfect NC = 10

The original image of size 256 times 256 and the watermarkof size 64 times 64

The watermarked image using middle frequencycoefficients (6 11 12 and 14 in Figure 2) with PSNR =31256 dB and NC = 10

options for precision hardware support for floating pointoperations and large on-chip RAM The CC++ compileris able to perform optimization (119899 = 0 1 2 and 3) indifferent level of clock cycles and code size 119899 = 0 the lowestlevel optimization provides the operations of performingloop rotation allocating variables to registers and simplifyingexpressions and statements 119899 = 1 the first level performsthe optimization on constant propagation and unused assign-ments 119899 = 2 the second level performs the optimizationon software pipelining unused global assignments loopunrolling and incremented pointer 119899 = 3 the highest

level performs the optimization by simplifying functionswithreturn values never used removing all functions never called119899 = 0 and 1 level can efficiently reduce the code size while119899 = 2 and 3 enhance the execution speed with larger codesize Higher levels generally result in the smallest code size

The code of the proposed watermarking method is writ-ten in C language After compiler assembler and linker thedevelopment software generates a machine file and stores itin DSPThe original image the watermark and the decodingkey are loaded and stored in SDRAM As a safety protectionin communication the decoding key as listed in Table 1


Table 3 Robustness of the watermarking method against image cropping and data compression

Image attack on watermarked image Retrieved watermark

5 cropping at the upper right area

NC = 0931

15 cropping from the upper right to the lower left

NC = 0777

14 data compression on the watermarked image

NC = 0707

is the protocol between the deliverer and the receiver Thecomputation time and code size in DSP at different levelsof optimization are listed in Table 5 The code size andcomputation time are 26KB and 089 seconds in the lowestlevel (119899 = 0) The computation time in first level (119899 =1) reduces to 061 seconds with code size 29KB In thesecond and third (119899 = 2 and 3) the computation timereduces to 035 and 033 seconds but the code size increasesto 43 and 44KB respectively Implementation on DSP ismuch faster than PC-based operation The higher the levelof optimization the less the time required By comparisonimplementation using Matlab on PC requires 16 secondsrsquocomputation time The experiment starts with the highestoptimization (119899 = 3) and the watermarked image andextracted watermark are shown in Table 5 with PSNR =31531 dB and NC = 0951 The calculation uses 57149130

clocks corresponding to 033 seconds in 44KB code size Inconclusion the watermarkingmethod onDSP is efficient andeffective for real-time implementation

5 Conclusions

(1) A digital image watermarking method based on thefast DCT algorithm is developed to maintain theimage quality and improve image robustness againstsignal processing The decoding keys include (i) thefour middle frequency coefficients (ii) two randompermutation vectors and (iii) a quantization matrixfor normalizing the watermark and the originalimage A watermark with ownership information isembedded into the host image in an unnoticeableway An index of peak signal to noise ratio (PSNR) is


Table 4 Watermarked images with watermark embedded in red green or blue component

Watermark embedded in the red component PSNR = 43156

Watermark embedded in the green component PSNR = 41251

Watermark embedded in the blue component PSNR = 43944

employed to measure the quality of the watermarkedimage and another the normalized correlation (NC)is also applied to evaluate the similarity and articu-lation of the extracted watermark with the reference(registered) watermark The ldquoblindrdquo watermarkingmethod is shown to be robust to image data croppingtransmission loss and compressiondecompression

(2) The watermarking method is first by decomposing atrue color image into RGB components such that a

watermark can be embedded in themain compositionfor better image quality Due to lower sensitivityof human eyes to blue light it is recommended toembed watermark in blue component if one cannotdetermine the main composition The image water-marking method is applicable to images in WindowsBitmap (BMP) Graphics Interchange Format (GIF)Joint Photographic Experts Group (JPEG) or TaggedImage File (TIF) format To maintain image quality


Table 5 Processing time comparison in DSP and in PC of the watermarking method

(a)

Optimization level Clocks (cycle) Time (s) Code size (KB)0 154645548 089 261 112044218 061 292 59152281 035 433 57149130 033 44PC with Matlab NA 160 NA

(b)

Optimization level 119899 = 3 Watermarked image Retrieved watermark

Four middle frequency coefficients forembedding watermark

PSNR = 31531

NC = 0951

after watermarking and to improve the robustness ofa watermarked image it is recommended to embedthe watermark in some of the middle frequencycoefficients (the 6thsim27th coefficients) in DCT asshown in the light color region of Figure 2 Selectionof the frequency coefficients should be based on thedesired watermarked image quality and robustnessFor an 8-bit original image of size 256 times 256 (119873 =256) and 8-bit digital watermark of size 64 times 64 (119872 =64) 4 middle frequency coefficients of the host imageafter DCT are needed to embed the watermark

(3) Experiments show that PC-based watermarking canbe completed within 16 seconds and the quality ofimage remains pristine with PSNR = 31256 dB andthe retrieved watermark NC = 10 The method isshown to be robust to image data cropping and com-pressiondecompression The method is also optimalfor real-time applications in DSP where implemen-tation shows that the embedding and extraction canbe completed within 033 seconds The watermarkingmethod is efficient and effective against illegal pirat-ing destruction and copy

Conflicts of Interest

The authors declare that there are no conflicts of interestregarding the publication of this paper

References

[1] A Koz and A A Alatan ldquoOblivious spatio-temporal water-marking of digital video by exploiting the human visual systemrdquoIEEE Transactions on Circuits and Systems for Video Technologyvol 18 no 3 pp 326ndash337 2008

[2] A Kaur R Dhir and G Sikka ldquoA new image steganographybased on first component alteration techniquerdquo InternationalJournal of Computer Science and Information Security vol 6 no3 pp 53ndash56 2009

[3] A Agarwal ldquoSecurity enhancement scheme for imagesteganography using S-DES techniquerdquo International Journalof Advanced Research in Computer Science and SoftwareEngineering vol 2 pp 164ndash169 2012

[4] V Tyagi A Kumar R Patel S Tyagi and S S Gangwar ldquoImagesteganography using least significant bit with cryptographyrdquoJournal of Global Research in Computer Science vol 3 no 3 pp53ndash55 2012

[5] J Nath S Das S Agarwal and A Nath ldquoA challenge in hidingencrypted message in LSB and LSB+1 bit positions in variouscover filesrdquo International Journal of Computer Applications vol14 no 7 pp 31ndash35 2011

[6] K Sara A D Mashallah and Y M Hossein ldquoA new steganog-raphy method based on HIOP (Higher Intensity Of Pixel) algo-rithm and strassenrsquos matrix multiplicationrdquo Journal of GlobalResearch in Computer Science vol 2 no 1 p 12 2011

[7] A Chadha N Satam R Sood and D Bade ldquoAn efficientmethod for image and audio steganography using Least Signif-icant Bit (LSB) substitutionrdquo International Journal of ComputerApplications vol 77 no 13 pp 37ndash45 2013


[8] H A Surrah and I N Mohamed ldquoA new steganography datahiding algorithmrdquo Journal of Global Research in ComputerScience vol 5 no 6 pp 1ndash6 2014

[9] X-Y Wang C-P Wang H-Y Yang and P-P Niu ldquoA robustblind color image watermarking in quaternion Fourier trans-form domainrdquo Journal of Systems and Software vol 86 no 2pp 255ndash277 2013

[10] A Benoraira K Benmahammed and N Boucenna ldquoBlindimage watermarking technique based on differential embed-ding in DWT and DCT domainsrdquo Eurasip Journal on Advancesin Signal Processing vol 2015 no 1 article 55 2015

[11] H M Al-Otum and N A Samara ldquoA robust blind colorimage watermarking based on wavelet-tree bit host differenceselectionrdquo Signal Processing vol 90 no 8 pp 2498ndash2512 2010

[12] C-S Lu S-K Huang C-J Sze and H-Y M Liao ldquoCocktailwatermarking for digital image protectionrdquo IEEE Transactionson Multimedia vol 2 no 4 pp 209ndash224 2000

[13] S Rawat and B Raman ldquoA blind watermarking algorithm basedon fractional Fourier transform and visual cryptographyrdquo SignalProcessing vol 92 no 6 pp 1480ndash1491 2012

[14] N MMakbol and B E Khoo ldquoRobust blind image watermark-ing scheme based on redundant discrete wavelet transform andsingular value decompositionrdquo AEUmdashInternational Journal ofElectronics and Communications vol 67 no 2 pp 102ndash112 2013

[15] S Lagzian M Soryani and M Fathy ldquoA new robust water-marking scheme based on RDWT-SVDrdquo International Journalof Intelligent Information Processing vol 2 no 1 pp 22ndash29 2011

[16] H-T Hu and L-Y Hsu ldquoExploring DWT-SVD-DCT featureparameters for robust multiple watermarking against JPEG andJPEG2000 compressionrdquo Computers and Electrical Engineeringvol 41 pp 52ndash63 2015

[17] A Mishra C Agarwal A Sharma and P Bedi ldquoOptimizedgray-scale image watermarkingusing DWT-SVD and fireflyalgorithmrdquo Expert Systems with Applications vol 41 no 17 pp7858ndash7867 2014

[18] S E Tsai and S M Yang ldquoA fast DCT algorithm for water-marking in digital signal processorrdquo Mathematical Problems inEngineering vol 2017 Article ID 7401845 7 pages 2017

Submit your manuscripts athttpswwwhindawicom

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

MathematicsJournal of


Mathematical Problems in Engineering

Hindawi Publishing Corporationhttpwwwhindawicom

Differential EquationsInternational Journal of

Volume 2014

Applied MathematicsJournal of


Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Journal of


Mathematical PhysicsAdvances in

Complex AnalysisJournal of


OptimizationJournal of


CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

International Journal of


Operations ResearchAdvances in

Journal of


Function Spaces

Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

International Journal of Mathematics and Mathematical Sciences


The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014


Algebra

Discrete Dynamics in Nature and Society



Decision SciencesAdvances in

Journal of


Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Stochastic AnalysisInternational Journal of


better robustness however both required substantial compu-tation with large data capacity

Singular value decomposition (SVD) is desirable forimplementation together with DCT or DWT because itrequires extensive computations if applied separately ontoimages Rawat and Raman [13] proposed an SVD-basedwatermarking scheme to increase security in copyright pro-tection Makbol and Khoo [14] and Lagzian et al [15] alsodeveloped image watermarking schemes based on DWT andSVD to achieve robustness against image processing but thecomputational load was substantial Hu and Hsu [16] pro-posed image watermarking with DWT-SVD-DCT featuresto guard against compression Mishra et al [17] applied afirefly algorithm based on DWT and SVD for watermarkingbut the encryption is time consuming The above SVD-based watermarking may have good robustness against JPEGcompression however the embedded watermark becomesvulnerable because all pixels would be changed when onesingular value is modified This paper presents an efficientdigital image watermarking method based on the fast DCTalgorithm for real-time application It is shown that thewatermarked image quality robustness and computationloading can all be improved

2 Embedding Watermark by the FastDCT Algorithm

Consider an 8-bit gray-level original image S and an imagewatermarkW

S = 119904 (119894 119895) 0 ≦ 119894 119895 ≦ 119873 minus 1 0 ≦ 119904 (119894 119895) ≦ 255 W

= 119908 (119901 119902) 0 ≦ 119901 119902 ≦ 119872 minus 1 0 ≦ 119908 (119901 119902) ≦ 255 (1)

where119872 and119873 are integers119872 lt 119873 The original image S isdivided into nonoverlapping pixel blocks of size 8times8 and thentransformed into frequency domain Each block is processedindependently by two-dimensional (2D) DCT into matrix FThe fast DCT algorithm [18] is applied to decompose the 2DDCT into a pair of 1D DCTs for computation efficiency Fora digital image S the 2D DCTmatrix F in frequency domainand its inverse transform back to spatial domain S1015840 can bewritten in matrix form by

F = M S M119879 (2a)

S1015840 = M119879 F M (2b)

where matrixM = 119898(119906 V) with 119898(119906 V) being the elementsin the 119906th row and Vth column

119898(119906 V)

=

12radic2 119906 = 0 cup 0 le V le 712 cos (2V + 1) 11990612058716 1 le 119906 le 7 cup 0 le V le 7

(3)

The 2D spatial datamatrix S1015840 in (2b) is the linear combinationof the base images which are obtained by the outer productof the column and row vectors of M By exploiting theredundancy in DCT coefficients it can be shown that the fastDCT algorithm can reduce the complexity of a 2D DCT of 8times 8 block from S to F in (2a) in only 24 multiplications andthe inverse transform from F to spatial domain S1015840 in (2b) issimply the transpose of the DCT matrixM [18]

Watermarking is performed first by decomposing a truecolor image into RGB components and the watermarkshould be embedded in the main composition of the orig-inal image to maintain image quality after watermarkingFor example an image rich in blue shall be watermarkedwith gray-level blue watermarks Due to the lower sen-sitivity of human eyes to blue light it is recommendedto embed watermark in blue component if one cannotdetermine the main composition of the image The imageembedding process of the watermarking method as shownin Figure 1 includes (1) computation optimization by thefast DCT algorithm and its inverse transform (2) selectionof the middle frequency coefficients (3) permutation ofthe watermark and (4) quantization (normalization) of thewatermark and the host image A spatial domain image S isprocessed by the fast DCT algorithm into spectral domainF Intuitively a watermark should be embedded into thehigher frequency coefficients in DCT for better perceptualinvisibility However the image ldquoenergyrdquo is concentrated inthe low frequency coefficients (eg the 0thsim5th coefficientsin Figure 2) so that the information embedded (hidden) inthe high frequency coefficients (the 28thsim63th coefficients)may be vulnerable to data processing Tomaintain the qualityof a host image after watermarking and to improve therobustness of a watermarked image against data processingit is recommended to embed the watermark in some of themiddle frequency coefficients (the 6thsim27th coefficients) asshown in the light color region of Figure 2 Because the hostimage of size 119873 times 119873 and the watermark119872 times119872 are dividedinto nonoverlapping pixel blocks of size 8 times 8 the number ofthe middle frequency coefficients selected for embedding thewatermark is (8119872119873)2 such that the sizes of the host imageand thewatermark are compatible For an 8-bit original imageof size 256times 256 (119873 = 256) and 8-bit digital watermark of size64 times 64 (119872 = 64) 4 middle frequency coefficients in each ofthe 8 times 8 blocks from the host image are needed to embedthe watermark The image block composed of these middlefrequency coefficients is denoted as Fm in frequency domain

A random number generator is applied to simultaneouslygenerate two vectors to permute the watermark W into Wpin spatial domain These vectors with elements in [1119872]representing the new coordinates (119909 119910) serve as a key towatermark decoding The permuted watermark is dividedinto nonoverlapping 4 times 4 blocks and processed by the fastDCT algorithm into Wf in spectral domain Quantization isrequired to normalize the image intensity of the host imageandwatermark thereby increasing the quality ofwatermarkedimage and reducing the watermarkmemory sizeThe processis to scale the watermark Wf into Wq by the host imageproperty represented by the average of the middle frequency


True color image

R B



G


Originalimage





Watermarked image



S F

W

Decomposition

FmFw

Fw

WpWf Wq

mdash

permutation

Sw



0 1 5 6 14 15 27 28

2 4 7 13 16 26 29 42

3 8 12 17 25 30 41 43

9 11 18 24 31 40 44 53

10 19 23 32 39 45 52 54

20 22 33 38 46 51 55 60

21 34 37 47 50 56 59 61

35 36 48 49 57 58 62 63



1198724minus1sum119894=0

1198724minus1sum119895=0

Wf (119886 + 4119894 119887 + 4119895)) (4)



PSNR = 10 log10 (2552MSE) (5)


MSE = 1647sum119894=0

7sum119895=0

(119904 (119894 119895) minus 1199041015840119908 (119894 119895))2 (6)











R

B


G



Registeredwatermark

Reverse permutation



Extractedwatermark


W

Decomposition

W











[44 19 34 9 62 7 50 57 39 20 40 30 10 49 53 27 52 43 6 13 4 26 23 47 55 3638 64 58 61 21 14 45 41 8 59 12 18 24 31 1 32 11 29 56 48 42 15 5 46 63 51

25 37 28 2 54 21 60 3 16 33 35 17][57 47 39 40 3 32 15 50 37 64 56 36 23 11 63 42 9 51 46 10 2 30 61 60 55 4143 59 26 25 27 4 35 45 20 53 7 12 8 31 21 44 62 5 33 16 54 6 22 18 14 17 13

1 19 49 29 52 58 38 34 28 48 24]


[[[[[[[[


]]]]]]]]











NC = 0931


NC = 0777


NC = 0707



5 Conclusions












(a)


(b)



PSNR = 31531

NC = 0951





References



























Volume 2014




Journal of











Journal of


Function Spaces






Algebra





Journal of





True color image

R B



G


Originalimage





Watermarked image



S F

W

Decomposition

FmFw

Fw

WpWf Wq

mdash

permutation

Sw



0 1 5 6 14 15 27 28

2 4 7 13 16 26 29 42

3 8 12 17 25 30 41 43

9 11 18 24 31 40 44 53

10 19 23 32 39 45 52 54

20 22 33 38 46 51 55 60

21 34 37 47 50 56 59 61

35 36 48 49 57 58 62 63



1198724minus1sum119894=0

1198724minus1sum119895=0

Wf (119886 + 4119894 119887 + 4119895)) (4)



PSNR = 10 log10 (2552MSE) (5)


MSE = 1647sum119894=0

7sum119895=0

(119904 (119894 119895) minus 1199041015840119908 (119894 119895))2 (6)











R

B


G



Registeredwatermark

Reverse permutation



Extractedwatermark


W

Decomposition

W











[44 19 34 9 62 7 50 57 39 20 40 30 10 49 53 27 52 43 6 13 4 26 23 47 55 3638 64 58 61 21 14 45 41 8 59 12 18 24 31 1 32 11 29 56 48 42 15 5 46 63 51

25 37 28 2 54 21 60 3 16 33 35 17][57 47 39 40 3 32 15 50 37 64 56 36 23 11 63 42 9 51 46 10 2 30 61 60 55 4143 59 26 25 27 4 35 45 20 53 7 12 8 31 21 44 62 5 33 16 54 6 22 18 14 17 13

1 19 49 29 52 58 38 34 28 48 24]


[[[[[[[[


]]]]]]]]











NC = 0931


NC = 0777


NC = 0707



5 Conclusions












(a)


(b)



PSNR = 31531

NC = 0951





References



























Volume 2014




Journal of











Journal of


Function Spaces






Algebra





Journal of





0 1 5 6 14 15 27 28

2 4 7 13 16 26 29 42

3 8 12 17 25 30 41 43

9 11 18 24 31 40 44 53

10 19 23 32 39 45 52 54

20 22 33 38 46 51 55 60

21 34 37 47 50 56 59 61

35 36 48 49 57 58 62 63



1198724minus1sum119894=0

1198724minus1sum119895=0

Wf (119886 + 4119894 119887 + 4119895)) (4)



PSNR = 10 log10 (2552MSE) (5)


MSE = 1647sum119894=0

7sum119895=0

(119904 (119894 119895) minus 1199041015840119908 (119894 119895))2 (6)











R

B


G



Registeredwatermark

Reverse permutation



Extractedwatermark


W

Decomposition

W











[44 19 34 9 62 7 50 57 39 20 40 30 10 49 53 27 52 43 6 13 4 26 23 47 55 3638 64 58 61 21 14 45 41 8 59 12 18 24 31 1 32 11 29 56 48 42 15 5 46 63 51

25 37 28 2 54 21 60 3 16 33 35 17][57 47 39 40 3 32 15 50 37 64 56 36 23 11 63 42 9 51 46 10 2 30 61 60 55 4143 59 26 25 27 4 35 45 20 53 7 12 8 31 21 44 62 5 33 16 54 6 22 18 14 17 13

1 19 49 29 52 58 38 34 28 48 24]


[[[[[[[[


]]]]]]]]











NC = 0931


NC = 0777


NC = 0707



5 Conclusions












(a)


(b)



PSNR = 31531

NC = 0951





References



























Volume 2014




Journal of











Journal of


Function Spaces






Algebra





Journal of





R

B


G



Registeredwatermark

Reverse permutation



Extractedwatermark


W

Decomposition

W











[44 19 34 9 62 7 50 57 39 20 40 30 10 49 53 27 52 43 6 13 4 26 23 47 55 3638 64 58 61 21 14 45 41 8 59 12 18 24 31 1 32 11 29 56 48 42 15 5 46 63 51

25 37 28 2 54 21 60 3 16 33 35 17][57 47 39 40 3 32 15 50 37 64 56 36 23 11 63 42 9 51 46 10 2 30 61 60 55 4143 59 26 25 27 4 35 45 20 53 7 12 8 31 21 44 62 5 33 16 54 6 22 18 14 17 13

1 19 49 29 52 58 38 34 28 48 24]


[[[[[[[[


]]]]]]]]











NC = 0931


NC = 0777


NC = 0707



5 Conclusions












(a)


(b)



PSNR = 31531

NC = 0951





References



























Volume 2014




Journal of











Journal of


Function Spaces






Algebra





Journal of








[44 19 34 9 62 7 50 57 39 20 40 30 10 49 53 27 52 43 6 13 4 26 23 47 55 3638 64 58 61 21 14 45 41 8 59 12 18 24 31 1 32 11 29 56 48 42 15 5 46 63 51

25 37 28 2 54 21 60 3 16 33 35 17][57 47 39 40 3 32 15 50 37 64 56 36 23 11 63 42 9 51 46 10 2 30 61 60 55 4143 59 26 25 27 4 35 45 20 53 7 12 8 31 21 44 62 5 33 16 54 6 22 18 14 17 13

1 19 49 29 52 58 38 34 28 48 24]


[[[[[[[[


]]]]]]]]











NC = 0931


NC = 0777


NC = 0707



5 Conclusions












(a)


(b)



PSNR = 31531

NC = 0951





References



























Volume 2014




Journal of











Journal of


Function Spaces






Algebra





Journal of








NC = 0931


NC = 0777


NC = 0707



5 Conclusions












(a)


(b)



PSNR = 31531

NC = 0951





References



























Volume 2014




Journal of











Journal of


Function Spaces






Algebra





Journal of














(a)


(b)



PSNR = 31531

NC = 0951





References



























Volume 2014




Journal of











Journal of


Function Spaces






Algebra





Journal of






(a)


(b)



PSNR = 31531

NC = 0951





References



























Volume 2014




Journal of











Journal of


Function Spaces






Algebra





Journal of























Volume 2014




Journal of











Journal of


Function Spaces






Algebra





Journal of











Volume 2014




Journal of











Journal of


Function Spaces






Algebra





Journal of




Documents

An Efficient Image Watermarking Method Based on Fast ...downloads.hindawi.com/journals/mpe/2017/3509258.pdfbut the encryption is time consuming. The above SVD-basedwatermarkingmayhavegoodrobustnessagainstJPEG