Multi-Watermarking Scheme for Copyright Protection ...

0 downloads 0 Views 2MB Size Report
Figure 3. Fragile Watermarking for Authentication [26]. • Semi-Fragile Watermark: is robust to a certain extent (i.e. threshold), and less sensitive to detect any.
Multi-Watermarking Scheme for Copyright Protection and Content Authentication of DubaiSat-1 Satellite Imagery Saeed AL-Mansoori1 and Alavi Kunhu2 Associate Image Processing Engineer, SIPAD Image Enhancement Section Emirates Institution for Advanced Science & Technology (EIAST), Dubai, UAE [email protected] 2 Electrical Engineer, ECE Department Khalifa University of Science, Technology and Research (KUSTAR), Sharjah, UAE 1

ABSTRACT A non-secure transmission channel is considered as a major challenge in remote sensing. The commercial value of satellite imagery and the sensitive information it contains led engineers to look for different means to secure the ownership of satellite imagery and preventing the illegal use of these resources. Therefore, a blind multi-watermarking scheme for copyright protection and image authentication is proposed. The multi-watermarking scheme is based on designing two back-to-back encoders. The first encoder embeds a robust ownership watermark in a frequency domain of satellite imagery using Discrete Cosine Transform (DCT) approach. Whereas, the second encoder embeds a fragile authentication information into a spatial domain of a watermarked image using Message Digest Encryption Key algorithm. This study was conducted on DubaiSat-1 satellite imagery owned by Emirates Institution for Advanced Science and Technology (EIAST). The simulation results demonstrate that the proposed scheme is robust against many intentional and unintentional attacks. Moreover, it shows a very high ability for tamper detection. Keywords: Multi-Watermarking, Copyright protection, Authentication, DubaiSat-1, Discrete Cosine Transform (DCT), Least Significant Bit (LSB), Index Mapping, Attack, JPEG Compression, Tampering.

1. INTRODUCTION

T

he rapid growth in the digital computer technology, and the wide use of digital documents, videos, audios and photos, has led to faster transmission of information, and easier distribution of data. Therefore, such technologies should offer a minimum level of protection to the intellectual property rights against misuses and illegal activities. Moreover, many recent events proved an increase in such illegal activities, and therefore demonstrated the importance of developing techniques to protect digital products. Encryption is a well-known procedure for securing data transmission [1]. The commonly used encryption methods are public key encryption (RSA) and Data Encryption Standard (DES) [1,2]. Such encryption methods achieve a certain level of security; nevertheless they make the secret message illegible and irregular. These irregular messages usually attract some unintended observers’ attention [1]. This led researchers to look for different means to secure the ownership of multimedia products and preventing the illegal use of these resources. One way for copyright protection and content authentication is through hiding a “digital signature” or “digital watermarking” within the product. Previously, different solutions were used to protect the copyright, such as the legal registration of the digital products with an authority concerned with copyright protection. However, it turned out more complicated compared to applying techniques such as watermarking. That method overwhelms the concerned authorities and requires them to keep a huge database to store all the registered information. The use of digital watermark technology, which has become a more convenient method for intellectual property protection, depends on hiding a mark or a signature in a digital medium. Usually, the watermark provides information about the copyright owner of this digital product. It could be a brand image, a serial number, or any other digital information that describes the owner. Recently, lots of studies [4-15] are focused on the development of robust watermarking. Such watermark is designed by embedding a perceptually transparent signal in the original signal in a way that the quality of the content will not be degraded, so as to resist any intentional or unintentional attacks leading to the destruction of the watermark. Therefore, robust watermarking is designed to protect the ownership of multimedia content. Satellite Data Compression, Communications, and Processing IX, edited by Bormin Huang, Antonio J. Plaza, Chein-I Chang, Proc. of SPIE Vol. 8871, 88710B © 2013 SPIE CCC code: 0277-786X/13/$18 · doi: 10.1117/12.2021855 Proc. of SPIE Vol. 8871 88710B-1 Downloaded From: http://proceedings.spiedigitallibrary.org/ on 01/06/2014 Terms of Use: http://spiedl.org/terms

Furthermore, there is otheer kind of waatermark whicch is equally important to o robust waterrmarking kno own as fragilee watermarkingg. This kind of o watermarkin ng is designedd by embedding an imperceeptible waterm mark in the orriginal contennt to detect and localize any minor m modificcation applied to the waterm marked conten nt. In this study a multi-waterm marking techn nique is designned based on a combination n of robust annd fragile wateermarking (i.ee. back-to-back encoders) in order to impllement a secuure tool which h leads to protect the owneership and autthentication of DubaiSat-1 ssatellite imagees, owned by Emirates Insttitution for Advanced A Scieence and Techhnology (EIA AST). The firsst encoder is deesigned by usiing Discrete Cosine C Transfform (DCT) teechnique, whiile Least Signnificant Bit (LSB) is used too design the second encoderr. The rest of the t paper is oorganized as follows. f Sectio on 2 briefly inntroduces the watermarkingg concept. In seection 3, the DCT D domain watermarking w g is discussed, whereas section 4 highlighhts the Least Significant S Biit (LSB) approaach. In sectionn 5, the propo osed method, D DCT-LSB waatermarking, is described inn detail. Sectio on 6 discussess the output of the methodoloogy and explaains the resultss. The conclussion of the wo ork is given inn section 7.

DCT Robu ust Encoderr

LSB Fragile Encoder

Original Im mage

r4

DCT T watermarked Im mage

Multi-wattermarked Imagee

Pattern Key

Watermark

Figure 1. Em mbedding Systeem Proposed

2. SEL LECTED B BACKGRO OUND MAT TERIAL 2.1. General Watermarkiing Framewo ork In general, a w watermarkingg scheme conssists of three pparts; the wateermark which is unique for each owner, the t embeddingg function whicch is called “eencoder”, and d the extractioon function which is the “d decoder”. Thee encoder is a secure part inn the watermarrking scheme,, which takes the Signal S (i.e. image) and the Wateermark Ѡ as an input and generates thee watermarked signal Sw (waatermarked im mage) using a specific key. This secret keey is used for encryption, which w offers ann additional prootection level.. Afterward, th he watermarkked signal is trransmitted through a comm munication chaannel, which iss considered ann insecure parrt where imag ge processing can play a role. Any imagee processing m manipulation, which affectss the quality off the retrieved watermark, iss called “Attacck”. Stirmark is a well-known image i manipu ulation tool thhat consists off 90 different types of imagge attacks thatt include lossyy compression (JPEG and MPEG), M geomeetric distortionns such as scaaling, and rotation attacks. The third and d main part inn watermarkingg scheme is thhe extraction function, f the ““decoder”. This part of the watermarking w g technique maay, or may noot be, secure deepending on thhe previous parts. p It is connsidered securre only if the embedding ffunction was robust againsst attacks and nnoise in the coommunication n channel. Finnally, the wateermark will be b extracted too prove the prroperty and too evaluate the pperformance of o the proposeed algorithm. F Figure 2 illusttrates the geneeral digital waatermarking sccheme. e/ public key

R

Privat

Attacks (1maige Processing Manipul atlon)

Siq

um ioeuumg

tiara

Fun ction F(E)

iel

Function

F4 S)

Waten

S

T Wutvermark (W)

sec :ure Part

Secure or Insecure Part

Figure F 2. Generral Digital Wateermarking Scheme

Proc. of SPIE Vol. 8871 88710B-2 Downloaded From: http://proceedings.spiedigitallibrary.org/ on 01/06/2014 Terms of Use: http://spiedl.org/terms

2.2. Watermarking Requirements A watermarking scheme must include the following properties: Transparency: The most essential requirement for any invisible watermarking technique. The watermark must be transparent to the end-user. There should not be any noticeable change between the original and the watermarked images. As the quality of remote sensing imagery is the main concern for end-users, the watermark content should not affect the image quality. • Security: One of the major properties of an efficient watermarking technique is being secure. This means that only the authorized parties can access them (i.e. owners). These parties can also change the content of the watermark. The encryption system should protect the watermark from unauthorized access to the watermarked content. • Ease of embedding and retrieval: When a watermark is uploaded on a digital media, it must be shown “on the fly” [1]. This requires having minimum computation for the chosen algorithm. • Robustness: Watermarking must be robust enough to protect the watermarked content against unauthorized access (i.e. attacks). In digital image processing, the term “Attack” is considered as any manipulation that alters the data content such as image rotation, resizing, filtering, cropping, compressing and adding noise. Thus, robustness of the algorithm is used as a key measure in evaluating the success of the watermarking technique. • Effect on bandwidth: The watermarking algorithm should produce watermarked content that does not increase the required bandwidth, compared to the bandwidth requirement of the actual content. •

2.3. Types of Watermarking Watermarking techniques can be categorized depending on the embedding domain, watermark type and the visibility of the watermark. For the embedding domain, watermarking techniques can be classified either as spatial (time) domain or transformed (frequency) domain. The spatial domain approach is less complex, while the frequency domain approaches are more robust and mostly used for image watermarking [3],[4]. Watermarking techniques are also categorized based on the watermark type, which can be classified as; text watermarking, image watermarking, audio watermarking, and video watermarking. Furthermore, the watermark can be either visible or invisible according to the human perception, where the invisible watermark can be further classified as: Robust Watermark: is a watermark which can be retrieved even if the watermarked image exposed to image processing manipulations (i.e. attacks). Such watermark is used in many applications with a view to protect the copyright of an image. • Fragile Watermark: is non-robust (i.e. sensitive) watermark, in since that any minor modifications applied to the watermarked image would destroy the watermark, which means that the watermark cannot be extracted and the copyright protection is lost. Such watermark is not recommended when transmitting a huge watermarked image using compression (i.e. unintentional attack) to an end-user. On the other hand, fragile watermark is designed to detect any change in pixel values and identify where it takes place in the image (i.e. tamper-proof). Thus, it is used to check the authenticity of an image. There are several techniques used to implement this kind of watermark, however, embedding the watermark in the Least Significant Bit (LSB) of the image is the most widely followed. •

Original Image

Modified Image

Modifications Detection

Figure 3. Fragile Watermarking for Authentication [26]

Semi-Fragile Watermark: is robust to a certain extent (i.e. threshold), and less sensitive to detect any changes in pixel values compared to fragile watermark.



Proc. of SPIE Vol. 8871 88710B-3 Downloaded From: http://proceedings.spiedigitallibrary.org/ on 01/06/2014 Terms of Use: http://spiedl.org/terms

In many digittal signal proccessing applications it is neccessary to guaarantee that th he sent signal ((i.e. image daata) is receivedd as it was sentt. Hence, any kind k of attack k that occurs too the signal sh hould be identtifiable. This lleads to securee the source of the transmitteed signal andd the signal authenticity. a Content Auth hentication caan be classifiied into two classes; exacct authenticationn and selectivve authentication. Exact authhentication deetects any slight change to tthe original signal (messagee or image) and rejects the received sign nal. On the oth ther hand, in selective auth hentication moodifications to o the signal iss allowed until a certain threeshold.

3. DISC CRETE CO OSINE TRA ANSFORM M (DCT) Discrete Cosiine Transform m (DCT) is ussed in a varieety of signal processing p app plications succh as data com mpression andd pattern recoggnition. It is a frequency linear l transfoormation dom main approach which is cha haracterized ass more robusst against attackks compared to t spatial (tim me) domain appproaches. Baasically, DCT is a process oof converting a signal from m spatial domaiin to its frequuency domain.. There are thr hree regions (ii.e. sub-bands)) of frequencyy the image will w be dividedd into; low freequency sub--band (FL), middle m frequuency sub-ban nd (FM) and high frequeency sub-band d (FH). DCT T watermarkingg can be classsified into blo ock based DC CT and global DCT. Transfforming an im mage via block based DCT T, will divide it into a non-ovverlapping × block, whhere( × ) components within the block ck are widely used. u Figure 4 shows a blocck of componnents illustratting the regioons of frequeency sub-band ds. The energgy is distribu uted from low w frequency to high frequenccy which means that from hhigh energy to o low energy in i a zigzag sccanning manneer as shown inn figure 5. Thee human visuaal system (HV VS) is more s ensitive to low w frequency sub-band coef efficients wherre most of thee signal energyy which contaains visually significant s infformation liess. This meanss that any moodification ap pplied to thosee coefficients w will cause disttortion to the host image. IIn addition, hiigh frequency coefficients aare considered d insignificannt because imagge processing manipulation n such as com mpression tend d to destroy high h frequencyy coefficientss aggressivelyy. Therefore, em mbedding wateermark in the middle frequeency sub-band d coefficients will not affecct the visibility y of the imagee as well as thee embedded watermark w willl not be destrroyed by com mpression. Thee output of DC CT is a real number, n unlikee the Discrete F Fourier Transsform (DFT) that t gives a coomplex number as an outpu ut. This is connsidered as an n advantage of DCT against DFT. On thee other hand, [7] show thatt the DCT is not robust ag gainst geomettric attacks su uch as scalingg, cropping, rotaation etc.

3M2

rrr rrr rrr rrr

L

F

IV

rrr rrr rrr

H Figure 4. DCT regions

Figure 5. Zigzaag scanning

Mathematicallly, the DCT transform t can be expressedd as follows [5 5]:

Where

αu annd α v

M −1 N −1 ( 2 m + 1) uπ ( 2 n + 1)vπ 2 α u α v ∑ ∑ x ( m, n ) coss cos 2M N 2N u =0 v=0

2 M

y (uu , v ) =

(1))

are giveen by:

⎧ 1/ 2 , u = 0

αu = ⎨

1



, u = 1,2,...., N − 1

,

⎧ 1/ 2 , v = 0

αv = ⎨ ⎩

1

, v = 1,2,...., N −1

(2))

Then, the imaage can be recconstructed by y applying invverse DCT bassed on equatio on (3);

x(m, n) =

2 M

2 M −1 N −1 (2m + 1)uπ (2n + 1)vπ ∑∑αuαv y(u, v) cos 2M cos 2N N u = 0 v =0

Proc. of SPIE Vol. 8871 88710B-4 Downloaded From: http://proceedings.spiedigitallibrary.org/ on 01/06/2014 Terms of Use: http://spiedl.org/terms

(3))

In equations (1), (2) and (3), x represents the original image having × pixels, ( , ) is the pixel intensity of an image in row and column and ( , ) is the DCT transform coefficients in row and column of the DCT matrix. Figure 6-b shows the effect of applying 2-dimensional DCT on a grayscale DubaiSat-1 image. The white color appears on the top-left corner of a DCT image, represents the location of the highest energy in an image.

( )

( )

Figure 6. ( ) Gray scale DubaiSat-1 image ( ) DCT representation of ( )

In order to study the DCT behavior, a random of ( × ) pixels are chosen from Figure 6-a. After applying a DCT, the coefficients will vary according to equations (1-3). As can be seen in figure 7, the highest energy coefficients in the image are concentrated on the top-left corner, where the highest energy coefficient is called a DC component while other coefficients stand for AC components. Most of the signal energy lies at low-frequencies sub-band, which contains the most important visual parts of the image [5]. In addition, the Human Visual System (HVS) is more sensitive to this region, thus a slight modification in its coefficients might incur severe distortion to the image.

300 200 100 0 -100 0

0 2

2 4

v

4 6

6 8

u

8

Figure 7. Frequency Map

In [6], the main steps in DCT block based watermarking are described as follows: [Step1]: Segment an image into non-overlapping blocks. [Step2]: Apply forward DCT to each of these blocks. [Step3]: Apply a block selection criteria based on the Human Visual System. [Step4]: Apply coefficient selection criteria (i.e. lowest, middle or highest). [Step5]: Embed a watermark by modifying the selected coefficients. [Step6]: Apply inverse DCT transform on each block. Steps 3 and 4 differ from one DCT block based algorithm and another because they are based on selection criteria.

Proc. of SPIE Vol. 8871 88710B-5 Downloaded From: http://proceedings.spiedigitallibrary.org/ on 01/06/2014 Terms of Use: http://spiedl.org/terms

4. LEAST SIGNIFICANT BIT (LSB) Embedding watermark in the Least Significant Bit (LSB) of the pixels is the initial work done in spatial domain digital image watermarking. This approach is simple and it is based on the idea that an image consists of pixels, and each pixel represented by 8 bits (i.e., pixel value 0 is represented as 00000000 in binary and pixel value 255 is represented as 11111111 in binary).The watermark information is first converted into binary format and then embedded in the least significant bit of selected pixels of the image as shown in figure 8. However, the difference in modulating the leastsignificant bit will not be observed by human eye due to the amount of vary is small. The disadvantage of using this method is that it is extremely vulnerable to attacks. Thus, any simple conversion in image formats may affect, and can destroy, the hidden information in the image. However, it can withstand simple operations such as addition of noise and cropping. 201 211

231 224

226 237

223 217

220 223 221 223 220

220

>

Decimal to Binary

225 251

11001001 11010011

11100111

11100000

11100010 11101101

11011111

11011001

11011100 11011111

11011101

11011100

11011111 11011100

11100001

11111011

Cover image (4.4 sob -block)

1

0

0

0

V

o

1

0

0

0

0

1

0

Least Significant Bit

0

0

0

1

(4.4 wat rmark)

11001001

11010100

11100010

11101101

11101000 11100000 11100000 11011010

11011100

11011111

11011101 11011100

11100000

11011100

11100010 11111011

Binary to Decimal

201 212

232 224

226 237

224 218 221 220 226 251

220 224

224 220

watermarkeaimage

Figure 8. An example of LSB approach

From literature survey, it is observed that many researchers applying LSB in designing their suggested embedding systems. For an instance, Abdullah Bamatraf et al. [8] proposed a LSB technique based on inversing the binary values of the watermark and shifting the watermark (text) consistent with the odd/even number of pixel coordinates of an image before embedding the watermark text. The proposed technique depends on the length of the watermark text. If the watermark’s length is more than ((M×N)/8)-2, the technique will embeds an extra watermark in the 2nd LSB. Moreover, this paper studying the effect of combing various LSB positions like 2nd LSB, 3rd LSB, 4th LSB and their combination. Abdullah Bamatraf et al. compared their simulation results with both the traditional LSB [9] and Lee’s technique [10]. It indicates that the proposed technique is robust against adding noise and cropping attacks. Recently, a novel technique is presented using the replacement of 2nd LSB with inverse of its corresponding LSB. Authors [11] are compared their recommended technique with simple LSB, DCT and DWT. A survey of LSB watermarking techniques is also available in [11]. Mohamed Ali Hajjaji et al. [12] proposed a new technique to embed a set of data (i.e. watermark) into medical images using LSB approach. Their technique is designed in a way to make the embedded data imperceptible and robust against attacks. This data contains the signature of the host image, the patient information (e.g., first name, family name, age, sex, etc.) and his diagnostic. The aim of this study is to confirm the integrity of the medical images and preserve the confidentiality of patient data.

Proc. of SPIE Vol. 8871 88710B-6 Downloaded From: http://proceedings.spiedigitallibrary.org/ on 01/06/2014 Terms of Use: http://spiedl.org/terms

5. PROPOSED METHODOLOGY 4bit Encryption Key

1 DCT Robust

Encoder Original Image

Index Mapping

' ... N

_

LSB Fragile

Encoder

DCT Watermarked Image

Multi-Watermarked arked Image

Pattern Key

Attack

EI ST Watermark

LSB Fragile

Decoder (Authentication)

DCT Robust

Decoder (Ownership) W

EIAST

Watermark

No Tampering

Tampered Image Any slight modification /tampering will be detected

Figure 9. A block diagram of the proposed system

As shown in figure 9 above, the proposed methodology is based on designing two back-to-back encoders. The first encoder is called “DCT Robust Encoder” , which provides a DCT watermarked image of ( × ) original color DubaiSat-1 satellite image using ( × ) color watermark (EIAST logo), and 4bit Encryption key. This DCT watermarked image will be passed through a second encoder called “LSB Fragile Encoder” and the message digest encryption algorithm is used to generate a unique 128 bit encryption key using ( × ) random selection pattern-key shown in figure 12. Thus, a multi-watermarked image will be produced which leads to increase the system security. In this study, the 16 index color watermark information is created using Adobe Photoshop software and each 24 bit color pixel of watermark logo is converted into 4 bit index information using index mapping table illustrated in Table 1. 5.1. Embedding Process: Encoder1: DCT Robust Encoder This encoder is designed to embed a ( × ) EIAST logo into ( × ) color DubaiSat-1 satellite imagery using Discrete Cosine Transform (DCT) approach. Such encoder provides robust watermarking which leads to protect the ownership of DubaiSat-1 satellite imagery against illegal violations. In our scheme, initially one channel is selected to be either Green (G) in case of RGB color space image or Luminance (Y) in case of YCbCr image. These selections are adopted based on the results of the study undertaken to determine the best RGB and YCbCr layers in terms of retrieving hidden information from a watermarked image, under unintentional attack (i.e. JPEG Compression). The results are shown in figures 10 and 11. The selected layer is divided into non-overlapping ( × ) sub-blocks. For each sub-block, ( , ), ( , ), ( , ), ( , ) corresponding to the highest DCT the best four coefficient locations amplitudes are obtained by applying DCT coefficient selection method. Afterward, the 4bit index information is embedded in the best coefficient locations using Odd/Even encoding method as illustrated in equations (4-5).

Proc. of SPIE Vol. 8871 88710B-7 Downloaded From: http://proceedings.spiedigitallibrary.org/ on 01/06/2014 Terms of Use: http://spiedl.org/terms

O k (u , v ) = DCT {o k (i , j )}, if w (i , j ) = 0 then ⎧ ⎛ O ( x, y ) ⎞ ⎟ ⎪ΔQ ⎜ k Ok ( x, y ) = ⎨ o ⎝ Δ ⎠ ⎪ O k (x, y ) ⎩

x, y ∈ Hk

1 ≤ k ≤ N HB

x, y ∉ Hk

1 ≤ k ≤ N HB

x, y ∈ H k

1 ≤ k ≤ N HB

x, y ∉ H k

1 ≤ k ≤ N HB

(4)

if w(i , j ) = 1 then ⎧ ⎛ O ( x, y ) ⎞ ⎪ΔQ ⎜ k ⎟ Ok ( x , y ) = ⎨ e ⎝ Δ ⎠ ⎪⎩ Ok ( x , y )

(5)

Where ( , ) represents ( × ) sub-block of the selected channel (green or luminance) of the image, ( , ) is its indicates the even quantization and DCT coefficients, and ( , ) represents the watermark. ≤ , ≤ and indicates the odd quantization to the nearest number. This symbol ‘Δ’ indicates the various scaling factor used for both quantization process. Compare Normalized Correlation (NC) for RGB Layers Watermarked Logo 1 0.95

Compare Normalized Correlation (NC) for YCbCr Layers Watermarked Logo 1

Red Layer Green Layer Blue Layer

0.9 Watermarked logo NC

Watermarked logo NC

0.9 0.85 0.8 0.75 0.7

0.85 0.8 0.75 0.7 0.65

0.65

75

Y Layer Cb Layer Cr Layer

0.95

0.6

80

85 90 JPEG Compression [%]

95

100

Figure 10. Comparing Normalized Correlation for RGB Layers Watermarked Logo under JPEG Compression Attack

0.55 75

80

85 90 JPEG Compression [%]

95

100

Figure 11. Comparing Normalized Correlation for YCbCr Layers Watermarked Logo under JPEG Compression

Encoder2: LSB Fragile Encoder This encoder is designed based on embedding a unique 128 bits message digest encryption key into a ( × ) DubaiSat-1 satellite imagery using Least Significant Bit (LSB) insertion method. Such encoder provides fragile watermarking which proves the source authenticity that has been subjected to potential tampering attacks. Fragile watermarking encoder takes the DCT watermarked image produced by Encoder 1 and divides it into non-overlapping ( × ) sub-blocks. Then, the sub-blocks pixels are divided into two groups called “key-group” and “LSB-group” using a unique pattern key as shown in figure 12, where white pixels represent key-group and black pixels represent LSB-group. The next step is to find the unique 128 bits encryption key for each key-group pixels using Message Digest Encryption Algorithm and embed 128 bits encryption key into LSB-group pixels using LSB insertion method. 5.2. Extraction Process: Decoder1: DCT Robust Decoder The DCT decoder is used to verify the owner of a multimedia object. In this paper, it is implemented to prove EIAST’s ownership of DubaiSat-1 satellite imagery by extracting EIAST logo information from multi-watermarked DubaiSat-1 imagery. The robust watermark is extracted by selecting the same layer (Green channel (G) of RGB image or Luminance (Y) of YCbCr image) used for DCT encoder, and divide it into non-overlapping ( × ) sub-blocks. Afterward, a 2DDCT is applied to transform these sub-blocks to the frequency domain. Then the encrypted 4 bits watermarked color logo index information is extracted from the best coefficient locations using Odd/Even extraction method as shown in equation 6 and decrypts it using decryption key.

Proc. of SPIE Vol. 8871 88710B-8 Downloaded From: http://proceedings.spiedigitallibrary.org/ on 01/06/2014 Terms of Use: http://spiedl.org/terms

⎛ O (u , v) ⎞ if Q ⎜ k ⎟ ⎝ Δ ⎠ ⇓ Even



Odd



w (i, j ) = 0 (66)

⇓ w (i, j ) = 1 In addition, the 4 bits inndex informattion is converrted into colo or logo using g the index m mapping tablee. In order too successfully eextract the watermark info ormation, bothh robust encoder and decod der should usse the same best coefficiennt locations, enccryption/decryyption key and d index mappiing table. Decoderr2: LSB Fragiile Decoder The next stepp is to extract the t fragile waatermark to chheck the imagee authenticity.. Thus, any sliight modificattion applied too the multi-wattermarked im mage will be detected. d The fragile waterrmarking deco oder first diviides the multii-watermarkedd image into nnon-overlappinng ( × ) sub-blocks and then re-divides the su ub-blocks pixxels into two groups calledd “key-group” and “LSB-grooup” using thee same uniquee pattern key used u at fragilee encoder. Thhen it extracts the embeddedd 128 bit encryyption key froom LSB-grou up using LSB B extraction method, m and compares it w with the messaage digest keyy generated froom key-groupps pixels. If both b keys aree different, it indicates thaat some tamppering has occcurred on thee ( × ) suub-block areaa of multi-wateermarked coloor DubaiSat-1 image.

rig Figure 122. Our unique paattern key Table 11. Index Mappin ng Table B Inilex Map )

55

54

0

0

255 255

2

1

[3

84 154

3

+5

24

5

201 245 154 73 224 178 107

6

70

2 14 >0

r2

6

4

7 8 9

10 11

12 13

)5

31 92

1

13

15

14

Proc. of SPIE Vol. 8871 88710B-9 Downloaded From: http://proceedings.spiedigitallibrary.org/ on 01/06/2014 Terms of Use: http://spiedl.org/terms

6. RESUL LTS AND ANALYSIS A S Our proposedd method waas compiled in n a MATLAB B environmen nt and tested over a largee database (~7 70 images) of ( × ) colored DubaiSat-1 D satellite imagees. The resultts indicate thaat our methodd is robust ag gainst variouss common imaage processingg attacks succh as rotationn, JPEG comp pression and resize attackss. In addition n, it illustratess a good perforrmance in prooving image’s authenticity. In this paper, we used six DubaiSat-1 D saample images for the testingg purposes; IM MG1, IMG2, IM MG3, IMG4, IMG5 I and IM MG6. Moreover, EIAST logo o that was useed as a waterm mark is shownn in figure 13. The Peak Siggnal-to-Noise Ratio R (PSNR)) and Structurre Similarity Index Measureement (SSIM)) were used too evaluate the pperformance of o the multi-w watermarked im mage. The PS SNR measuress the level of nnoise in the prroduced signaal compared to tthe host signaal. The internaational unit useed to measuree PSNR is the decibel [dB].. Mathematicaally, the PSNR R is defined as:

PSNR = 10 × log10

22552 1 M ×N

(7))

∑ ∑ ( X (i, j ) − Y (i, j )) M −1

N −1

i =0

j =0

2

w of the innput image. Thus, T × represents r the host image siize. Moreoverr, Where, annd denote thhe height and width ( , ) repressents the originnal image, wh hereas, ( , ) indicates thee multi-waterm marked image.. In SSIM indeex, the similarrity measurem ment is based oon the structurral informatio on of the imagge. Therefore, and havee three levels oof comparisonns defined as: luminance ( )), contrast ( ) and structurral ( ) compaarisons. Matheematically, thee following term ms are definedd as follows: l( X ,Y ) =

2μ X μ Y + A μ X2 + μ Y2 + A

,

c (X ,Y ) =

2σ X σ Y + B σ X2 + σ Y2 + B

,

s ( X ,Y ) =

σ XX Y + C σ XX σ Y + C

(8, 9, 10))

Where and represent thhe local mean n intensity andd local standarrd deviation of the image. T The symbols, , and aree is the loocal covariance coefficient bbetween an nd . Thus, thee SSIM index between two images andd constants andd is given in equation (11). α SSIIM ( X , Y ) = [l ( X , Y ) ] . [ c ( X , Y )] β . [ s ( X ,Y )]γ (11)) Mainly, the S SSIM values are a ranging from 0 to 1. Thhus, the lowesst SSIM valuee is 0 which m means that different imagges while the highest SSIM M value is 1 whhich means th hat = . S

IMG1

IMG2

IMG5

r

IMG3

IMG6

IMG4

A waterrmark (128×1128 pixels)

Figure 13. 1 DubaiSat-1 sample imagess and EIAST log go (watermark)) used for testinng purpose

Proc. of SPIE Vol. 8871 88710B-10 Downloaded From: http://proceedings.spiedigitallibrary.org/ on 01/06/2014 Terms of Use: http://spiedl.org/terms

and

are twoo

From the simulation results and by Human Visual System (HVS), no observable difference between multi-watermarked and original images was noticed. Furthermore, the multi-watermarked image was assessed using image quality parameters (PSNR and SSIM) to evaluate its performance. The best range commonly considered for PSNR is between 40 – 55 dB [18]. Table 2 below illustrates the PSNR and SSIM for watermarked image provided from DCT Encoder, and the multi-watermarked image provided from the LSB Encoder using RGB format DubaiSat-1 sample images. Whereas table 3 illustrates the same assessment but using YCbCr image format. It is observed that the PSNR and SSIM are higher in case of watermarked image (DCT Encoder) compared to multi-watermarked image (LSB Encoder), which is logical. In addition, it is noticed that the PSNR and SSIM of DCT watermarked and multi-watermarked images are much higher in case of RGB image format. Table 2. PSNR and SSIM for DCT Watermarked and Multi-Watermarked RGB images DCT ENCODER PSNR SSIM

DCT+LSB ENCODER PSNR SSIM

IMG1 IMG2

43.74 dB 43.27 dB

0.994 0.999

43.37 dB 42.95 dB

0.994 0.999

IMG3 IMG4

43.99 dB 43.42 dB

0.996 0.998

43.59 dB 43.07 dB

0.995 0.997

IMG5 IMG6

43.30 dB 44.52 dB

0.999 0.999

42.98 dB 43.99 dB

0.999 0.998

Table 3. PSNR and SSIM for DCT Watermarked and Multi-Watermarked YCbCr images DCT ENCODER PSNR SSIM

DCT+LSB ENCODER PSNR SSIM

IMG1 IMG2

37.40 dB 37.06 dB

0.973 0.996

37.12 dB 36.81 dB

0.972 0.996

IMG3 IMG4

37.58 dB 37.20 dB

0.979 0.990

37.29 dB 36.93 dB

0.978 0.989

IMG5 IMG6

37.18 dB 38.01 dB

0.981 0.999

36.98 dB 37.89 dB

0.988 0.995

The next step is to analyze the system performance in terms of robustness and sensitivity to any tampering and modifications. To check whether the proposed method protects the ownership of DubaiSat-1 imagery, a couple of attacks are applied to test the system like rotation, JPEG Compression and Resize attacks. Therefore, the Normalized Correlation (NC) is used to designate the presence/absence of the hidden watermark. R

NC =

C

∑∑ (w i =1 j =1

i, j

⊕ wi`, j )

(12) × 100% R×C Where , represents a non-altered pixel color of extracted watermark from the original image, ` , represents an altered pixel color of extracted watermark from the original image and × is the watermark size. Thus, in this section a multi-watermarked IMG2 is selected to test the system robustness under three common attacks: (a) Rotation Attack In digital multimedia images, “Rotation Attack” is considered as the most popular kind of geometrical attack. The concept behind this attack is to rotate the multi-watermarked image clockwise by a certain degree. This will affect the watermark within the image. So, as the algorithm proposed was more robust and secure, the watermark will be retrieved with better quality. In this paper, five levels of rotations were implemented. The multi-watermarked image was rotated by 1 degree, 3 degrees, 5 degrees, 7 degrees and 9 degrees, respectively. Figures 14 and 15 demonstrate the Normalized Correlation (NC) analysis of recovered logo under various rotation degrees from G / Y channel of multi-watermarked DubaiSat-1 image (IMG2). Simulation results show that as the rotation degree increases the normalized correlation between the retrieved and original logos decreases. Furthermore, it is noticed that both image formats (RGB and YCbCr) gives similar correlation values.

Proc. of SPIE Vol. 8871 88710B-11 Downloaded From: http://proceedings.spiedigitallibrary.org/ on 01/06/2014 Terms of Use: http://spiedl.org/terms

1o

3o

5o

7o

9o

NC = 0.98159

NC = 0.95747

NC = 0.93254

NC = 0.91239

NC = 0.88619

Figure 14. Recovered logo under Rotation attacks from G channel of multi-watermarked color DubaiSat-1 image IMG2

1o

3o

5o

7o

9o

NC = 0.98343

NC = 0.96205

NC = 0.93949

NC = 0.92122

NC = 0.89932

Figure 15. Recovered logo under Rotation attacks from Y channel of multi-watermarked color DubaiSat-1 image IMG2

For RGB Image For YCbCr Image 1

NC

0.8

0.6

0.4

0.2

0

1

1.5

2

2.5

3

3.5

4

4.5

5

Figure 16. NC analysis of recovered logo under Rotation attacks from multi-watermarked color DubaiSat-1 image IMG2

(b) JPEG Compression Attack This attack is considered as one of the most severe attacks on the watermarked images. Typically this type of attack occurs unintentionally, for example if we need to send a large image to the client through the internet where the bandwidth is limited; we should compress the image in order to be received by the end-user. If the proposed method is robust, the embedded watermark will not be affected much. On the other hand, if the watermark is not robust, the watermark will be destroyed and the image’s ownership will be lost. To test the robustness of our proposed method against this attack, five levels of JPEG compression Quality Factors were implemented. Thus, the multi-watermarked image (IMG2) was compressed by 100%, 90%, 80%, 70% and 60%.

Proc. of SPIE Vol. 8871 88710B-12 Downloaded From: http://proceedings.spiedigitallibrary.org/ on 01/06/2014 Terms of Use: http://spiedl.org/terms

Figures 17 and 18 illustrate that our method is robust against JPEG compression attack especially when the image is in YCbCr format. The simulated results show that the normalized correlation between the retrieved and original watermark is much better in case of YCbCr compared to RGB where the watermark is destroyed and cannot be retrieved in RGB format while it is reasonable and can be retrieved in YCbCr case. Considering the case when compression Quality Factor is 70%, the NC of RGB format is 0.6543, while it is 0.9856 in case of YCbCr format.

100%

90 %

80 %

70 %

60 %

NC = 0.99175

NC = 0.86878

NC = 0.7921

NC = 0.6543

NC = 0.80293

Figure 17. Recovered logo under JPEG Compression attacks from G channel of multi-watermarked color DubaiSat-1 image IMG2

100%

90 %

80 %

70 %

60 %

NC = 0.99703

NC = 0.99479

NC = 0.99107

NC = 0.9856

NC = 0.97387

Figure 18. Recovered logo under JPEG Compression attacks from Y channel of multi-watermarked color DubaiSat-1 image IMG2 For RGB Image For YCbCr Image 1

NC

0.8

0.6

0.4

0.2

0

1

1.5

2

2.5

3

3.5

4

4.5

5

Figure 19. NC analysis of recovered logo under JPEG Compression attacks from multi-watermarked color DubaiSat-1 image IMG2

(c) Resize Attack Most of the time when posting a picture online or sending it via email, resize the picture may be necessary. It is a nontrivial process that involves a trade-off between efficiency, smoothness and sharpness. In the case of resize attacks, we are basically changing the size of the original image. That means either increasing or decreasing the total number of

Proc. of SPIE Vol. 8871 88710B-13 Downloaded From: http://proceedings.spiedigitallibrary.org/ on 01/06/2014 Terms of Use: http://spiedl.org/terms

pixels in the image, and then trying to recover the hidden information from the resized image. From simulation results, figures 20 and 21 illustrate that there is a minor preference in the normalized correlation of YCbCr image format compared to RGB format. For an instance, by resizing the image by 50%, NC for YCbCr is 0.90777 while NC for RGB is 0.86342.

50%

70 %

90 %

120%

160%

NC = 0.86342

NC = 0.92028

NC = 0.96701

NC = 0.9842

NC = 0.98446

Figure 20. Recovered logo under Resize attacks from G channel of multi-watermarked color DubaiSat-1 image IMG2

50%

70 %

90 %

120%

160%

NC = 0.90777

NC = 0.94766

NC = 0.97881

NC = 0.98968

NC = 0.99028

Figure 21. Recovered logo under Resize attacks from Y channel of multi-watermarked color DubaiSat-1 image IMG2 For RGB Image For YCbCr Image 1

NC

0.8

0.6

0.4

0.2

0

1

1.5

2

2.5

3

3.5

4

4.5

5

5.5

6

Figure 22. NC analysis of recovered logo under Resize attacks from multi-watermarked color DubaiSat-1 image IMG2

Proc. of SPIE Vol. 8871 88710B-14 Downloaded From: http://proceedings.spiedigitallibrary.org/ on 01/06/2014 Terms of Use: http://spiedl.org/terms

The Proposed LSB Fragile system (encoder and decoder) is designed to check the authenticity of our multi-watermarked DubaiSat-1 imagery. The method is very sensitive to any modifications or tampering. Thus, any slight modification on multi-watermarked satellite image will make a difference between the generated 128 bit message digest encryption key and LSB method recovered encryption key on all the tampered ( × ) sub-blocks. Our algorithm accurately shows all tampered areas on a colored multi-watermarked DubaiSat-1 image with a resolution accuracy of ( × ) pixel, and all non-tampered area are represented by black. To test our proposed fragile system, a couple of modifications are added to IMG2 using Adobe Photoshop software. Figure 23 shows that the proposed method is very sensitive to any tampering. t Multi-watermarked DubaiSat-1 Image (IMG2)

gib Case1: Tampered Multi-watermarked DS1 Image

Case2: Tampered Multi-watermarked DS1 Image

Tampered area detected for Case1

Tampered area detected for Case2

* Case3: Tampered Multi-watermarked DS1 Image

Tampered area detected for Case3

Figure 23. Various tampered (modified) area detected on multi-watermarked DubaiSat-1 satellite image

Proc. of SPIE Vol. 8871 88710B-15 Downloaded From: http://proceedings.spiedigitallibrary.org/ on 01/06/2014 Terms of Use: http://spiedl.org/terms

7. CONCLUSION In this paper, a multi-watermarking scheme is designed based on embedding a robust ownership watermark into DubaiSat-1 satellite imagery using Discrete Cosine Transform (DCT) and embeds a fragile authentication information using Least Significant Bit (LSB). The proposed scheme does not require an original image to extract the embedded information (i.e. watermark).The aim behind proposing such technique is to protect the intellectual property rights against misuses and illegal activities. In "DCT robust encoder", the robust watermark is embedded in either green channel if satellite imagery is in RGB format or Luminance in case of YCbCr format. The message digest encryption algorithm is used in "LSB fragile encoder" to generate a unique 128 bit encryption key using ( × ) random selection key. To verify the ownership of satellite imagery, "DCT robust decoder" is used to extract the embedded watermark and know the actual owner. Whereas, the "LSB fragile decoder" is used to check the image authenticity. From the simulation results, no observable difference between multi-watermarked and original image was noticed. Moreover, the PSNR and SSIM of DCT watermarked and multi-watermarked images are much higher in case of RGB image compared to YCbCr. Our proposed method is efficient in a since that it is robust against various attacks such as rotation, JPEG compression and resize attacks. Also, any slight modification or tampering attempt to multi-watermarked image was detected.

8. REFERENCES [1]Aruna Mittal, “A Highly Secure Skin Tone Based Optimal Parity Assignment Steganographic Scheme Using Double Density Discrete Wavelet Transform”, International Journal of Advanced Research in Computer and Communication Engineering, Vol.1,Issue 9, November 2012. [2]Mazhar Tayel, Hamed Shawky and Alaa El-Din Sayed Hafez, “A Hybrid Chaos-Fuzzy-Threshold Steganography Algorithm for Hiding Secure Data”, International Conference of Advanced Communication Technology (ICACT), 2013. [3] Saraju P. Mohanty , "Digital Watermarking: A Tutorial Review", 1999. [4] I.J. Cox, J.Kilian, T.Leighton and T. Shamoon, “Secure Spread Spectrum watermarking for Multimedia,” IEEE Tras. on Image Processing , Vol. 6,No12, 1997, pp. 1673-1687. [5]Ali AL-Haj, “Combined DWT-DCT Digital Image Watermarking”, Journal of Computer Science 3 (9):740-746, 2007. [6]Tribhuwan,Vikas, “ An Improved and Robust DCT Digital Image Watermarking Scheme”, International Journal of Computer Applications, Volume3-No.1,June 2010. [7]Manpreet, Sonika, Sunny, “A STUDY OF DIGITAL IMAGE WATERMARKING”, International Journal of Research in Engineering & Applied Sciences, Volume 2, Issue 2, February 2012. [8]Abdullah Bamatraf, Rosziati Ibrahim and Mohd.Najib, “A New Digital Watermarking Algorithm Using Combination of Least Significant Bit (LSB) and Inverse Bit”, Journal of Computing, Volume3, Issue 4, April 2011. [9]Kurah, C.AND Mchughes, J.1992. “A cautionary note on image downgrading”, In Proceedings of the IEEE Computer Security Applications Conference. Vol.2. IEEE Computer Society Press, Los Alamitos, CA, 153-159. [10]Lee, G.J., Yoon, E.J. and Yoo, K.Y. (2008), “A new LSB based Digital Watermarking Scheme with Random Mapping Function”, in 2008 IEEE DOI 10.1109/UMC.2008.33. [11] Amit Singh, Susheel Jain and Anurag Jain, “Digital Watermarking Method Using Replacement of Second Least Significant Bit (LSB) with Inverse of LSB”, International Journal of Emerging Technology and Advanced Engineering, Vol.3,Issue 2, February 2013. [12] Mohamed Ali HAJJAJI, Abdellatif MTIBAA and El-bey BOURENNANE, “A Watermarking of Medical Image: Method Based “LSB””, Journal of Emerging Trends in Computing and Information Sciences, Vol.2, No.12, December 2011. [13]G.Langelaar, I.Setyawan,R.L.Lagendijk, “Watermarking Digital Image and Video Data”, in IEEE Signal Processing Magazine, Vol. 17, pp 20-43, September 2000. [14]Baisa L.Gunjal, R.R.Manthalkar,”An Overview of Transform Domain Robust Digital Image Watermarking Algorithms”, Journal of Emerging Trends in Computing and Information Sciences, Vol.2 No.1, 2010. [15]Andreja, Jan, “Attacks on Digital Wavelet Image Watermarks”, Journal of Electrical Engineering, Vol.59, No.3, pp.131-138, 2008. [16]Keshav S Rawat,Dheerendra, “Digital Watermarking Schemes for Authorization Against Copying or Piracy of Color Images”, Indian Journal of Computer Science and Engineering, Vol.1, No.4, pp. 295-300.

Proc. of SPIE Vol. 8871 88710B-16 Downloaded From: http://proceedings.spiedigitallibrary.org/ on 01/06/2014 Terms of Use: http://spiedl.org/terms

[17]K. R. Rao and P. Yip, “Discrete Cosine Transform: Properties, Algorithms, Advantages, Applications”, Academic Press, Massachusetts, 1990. [18]Saeed AL-Mansoori, Alavi Kunhu, “Robust Watermarking Technique based on DCT to Protect the Ownership of DubaiSat-1 Images against Attacks”, IJCSNS International Journal of Computer Science and Network Security, VOL.12 No.6, June 2012. [19]S. K. Bandyopadhyay, D. Bhattacharyya, P. Das, “Quantum Watermarking and Extraction for handwritten signature”, Industrial Electronics and Applications, ICIEA 2008. 3rd IEEE Conference, pp. 959-963, 2008. [20]C. M. Kung, J. H. Jeng, and C. H. Kung, "Watermarking Base on Block Property," 16th IPPR Conference on Computer Vision, Graphics and Image Processing , pp.540-546, 2003. [21]C. M. Kung, T. K. Truong and J. H. Jeng, “A Robust Watermarking for Image Authentication Technique”, 2003 IEEE International CARNAHAN Conference on Security Technology, 37th Annual Conference, pp.400-404, 2003. [22]A. Sinha, A. Das, and S. Pandith, “Pattern based robust digital watermarking scheme for images,” Acoustics, Speech, and Signal Processing, 2002 IEEE International Conference on, vol. 4, pp.3481-3484 ,2002. [23]Saeed AL-Mansoori, “An efficient watermarking technique for satellite images using discrete cosine transform”, in High-Performance Computing in Remote Sensing II, Proceedings of SPIE Vol. 8539 (SPIE, Bellingham, WA 2012), 85390A. [24]D. Kundur, D. Hatzinakos, “A Robust Digital Image Watermarking Method Using Wavelet-based Fusion,” Proc. ICIP’97, vol. 1, pp. 544-547, 1997. [25]A. Zandi, E. L. Schwartz, and M. Boliek, “CREW: Compression with Reversible Embedded Wavelets,” Proc. Data Compression Conference, pp.212-221, 1995. [26] http://homepages.vub.ac.be/~andooms/research.htm. [27]A. Kunhu and H. Al-Ahmad, “Multi Watermarking Algorithm Based on DCT and Hash Functions for Color Satellite Images”, International Conference on Innovations in Information Technology, March 2013, Al Ain, UAE, pp. 30-35. [28] A. Kunhu and H. Al-Ahmad “A New Watermarking Algorithm for Color Satellite Images Using Color Logos and Hash Functions”, International Conference on Computational Intelligence, Communication Systems and Networks, June 2013, Madrid, Spain, pp. 251-255.

Proc. of SPIE Vol. 8871 88710B-17 Downloaded From: http://proceedings.spiedigitallibrary.org/ on 01/06/2014 Terms of Use: http://spiedl.org/terms