Performance Comparison of Product Codes and Cubic Product Codes using FRBS for Robust Watermarking

Transcription

1 Internatonal Journal of Computer Informaton Systems and Industral Management Applcatons. ISSN Volume 8 (206) pp MIR Labs, Performance Comparson of Product Codes and Cubc Product Codes usng FRBS for Robust Watermarkng Atta-ur-Rahman, Muhammad Azam 2 and Gohar Zaman Baran Insttute of Informaton Techology (BIIT), PMAS-AA Unversty, Rawalpnd, 46000, Punjab, Pakstan atta@bt.edu.pk 2 Baran Insttute of Informaton Techology (BIIT), PMAS-AA Unversty, Rawalpnd, 46000, Punjab, Pakstan mazam770@yahoo.com The Unversty of Lahore, Blue Area, Islamabad, 44000, Pakstan gohar.zaman.sb@gmal.com Abstract: In ths paper, a novel technque for robust dgtal mage watermarkng s proposed usng smple product codes (PC) and cubc product codes (CPC) respectvely. Product codes are two dmentonal matrx codes whle CPC are three dmensonal product codes where the consttuent codes n each dmenson are lnear block codes. The structure of PC and CPC makes them sutable for the proposed scheme. The embedded watermark s vulnerable to varous attacks on the mage lke compresson, nose and geometrc attacks namely translaton rotaton and scalng (TRS) etc. Ths could lmt the performance of dgtal watermarkng schemes. Our proposal s to encode the watermark wth PC/CPC pror to embeddng t nto the mage. Ths could easly be done because our watermark s also a three dmensonal data (mage/logo etc), so each dmenson can easly be encoded wth correspondng codes n PC/CPC. The modfed teratve decodng algorthm (MIDA) s employed to decode the PC/CPC. Moreover, a Fuzzy Rule Based System (FRBS) s used to fnd the sutable regons n the host mage where watermark can be embedded such that the mperceptbltyof the host mage should not be affected. The FRBS makes used of Human Vsual System (HVS) parameters to obtan those regons. The scheme s tested aganst varous attacks and compared wth the well-known schemes n the lterature. Moreover, role of PC s also compared wth CPCs and results are demonstrated. Keywords: Dgtal Watermarkng, Product Codes, Cubc Product Codes, Robustness, lnear block codes, MIDA, FRBS I. Introducton Dgtal watermarkng for authentcaton and copyrght protecton s one of the nterestng areas of research n nformaton securty. Ths technque s used for authentcaton, copyrght protecton, owner dentfcaton and copy control etc of a dgtal document. Ths technque s no more lmted to the mages but also appled to the audo, vdeo, softwares and databases etc. There are three basc propertes of dgtal watermarkng namely, capacty, mperceptblty and robustness. Capacty s measured wth the sze of watermark beng embedded. Imperceptblty s measured wth the degradaton that s caused by embedded watermark n the mage. That s the degradaton should not be notceabe. Robustness measures s resstance to certan attacks on the watermarked mage. Accordng to the watermarkng termnology, an attack s an event that can cause tamperng n the mage, thus makng the watermark dffcult to detect. Manly attacks can be dvded nto two categores, ncdental and malcous. Incdental attacks are frendly attacks and are requred sometmes for example, JPEG compresson s used n many of nternet applcatons to make the fle sze small. Malcous attacks can be dvded nto four man categores namely, geometrcal attacks, removal attacks, protocol attacks and cryptographc attacks as dscussed by Gokozan []. Product codes are serally concatenated codes and were ntally proposed by Elas [2] n 952. Product codes are two dmensonal lnear block codes. Later the same concept was extended to three dmensonal codes, called Cubc Product Codes (CPC) by []. In ths concept long codes were generated by usng much shorter consttuent block codes n each dmenson. Constructon process of the CPC s presented n a subsequent secton as well. Robust watermarkng schemes allow both ncdental and malcous attacks whle the fragle watermarkng schemes donot allow any modfcatons. Sem-fragle watermarkng schemes n such a way that they are robust aganst frendly modfcatons but are fragle aganst malcous attacks. Dynamc Publshers, Inc., USA

2 58 Mostly, retransmsson s consdered as a soluton to ths problem but only after the tamperng s detected (a case of fragle watermarkng). But when the tme s strngent then retransmsson may be costly (onlne scenaro) also there s no guarantee that the receved sgnals are error free. Morevoer, n retransmsson throughput s also compromsed. Smlarly use of cryptography for securty s always a good choce. But the property that makes a cpher strong, makes t senstve to the channel error at the same tme. Soluton s agan retransmsson but at the cost of throughput [4]. A relable wreless error correcton technque for secure mage transmsson s proposed n [5], where turbo codes were used for error free communcaton n contrast to chaos based encrypton technque. Real BCH (Bose Choudhary Hoqagan) codes have been nvestgated for robust mage transmsson usng a jont source-channel codng technque [6]. Error Correctng Codes (ECC) provdes error free communcaton at the cost of redundancy. There are two major types of ECC that s Convolutonal Codes (CC) and Lnear Block Codes (LBC) [7]. In [8], authors proposed a reversble watermarkng technque that mproves the securty of medcal mages wth addtonal features to detect the tamperng regon and then to recover the tamperng regon of the watermarked mage. A Resdue Number System (RNS) based reversble watermarkng was proposed n [9]. In ths paper authors used RNS to rescue the watermark. The proposed scheme was hghly fragle aganst all knd of attacks. In [0], authors proposed a robust watermarkng scheme resstant to geometrc attacks. Ths was accomplshed by Arnold s transform. It was also stated that transformed doman technques provde sgnfcant mprovement n robustness. Atta-ur-Rahman et al [] proposed a novel technque for relable mage transmsson usng Product Codes. In that technque the mage was encoded pror to transmsson. Product codes beng comprsed of two dmensonal block codes, were observed structurally compatble wth the mages. Atta-ur-Rahman et al [2] proposed a novel robust watermarkng technque usng Cubc Product Codes and Fuzzy Rule Based System. In ths technque the watermark was encoded by CPC pror to embeddng n the host mage. FRBS was used to embed the watermark on sutable places n the host mage. The scheme was demonstrated robust aganst a number of attacks. In ths paper, PC and CPCs are employed to dgtal mage watermarkng for robustness. A specfc szed watermark (mage) s encoded by PC/CPC, pror to embeddng nto the mage. The embeddng postons (pxels) for the watermark are obtaned from a fuzzy rule based system (FRBS) that hghlghts the postons ntutvely on behalf of parameteres of HVS. The scheme s verfed for natural as well medcal mages. Rest of the paper s organzed as follows: Secton 2 presents the constructon of PC and CPC and ther decodng technque, the fuzzy rule based system s gven n Secton, watermark embeddng and extracton s dscussed n Secton 4, results of the proposed scheme are depcted n secton 5 whle secton 6 concludes the paper. II. Constructon of PCs and CPCs Rahman et al. Ths secton contans the bref ntroducton to the constructon of product codes and cubc product codes respectvely. Cubc Product Codes are an extenton to smple PC. A. Product Codes Product codes are serally concatenated codes n whch short consttuent codes are used to construct bgger codes. In Product codes that are two dmentonal (matrx) codes are conssted of lnear block codes n row and column wse. Ths arrangement s shown n fg-. Consder two lnear block codes A and A2 wth the codng parameters [ n, k, d] and [ n2, k 2, d 2] respectvely, where n, k and d ;, 2 are the length, dmenson and mnmum Hammng dstance ( d mn ) of the code A (,2) respectvely. Code A wll be used as row code whle A2 wll be used as column code. The rates of ndvdual codes are R and R respectvely gven by, 2 k R,,2 () n The product code Ω can be obtaned by codes the followng manner. Place and k k 2 k columns k 2 nformaton bts n an array of Encode rows usng code an array of k2 n Now encode result n n2 nproduct code. Fgure. A n columns usng code A,,2 k 2 n rows, whch wll result n A 2 Structure of the Product code, whch wll The resultant product code Ω has the parameters [ nn2, kk 2, dd 2] and the rate wll be RR 2. In ths way long block codes can be constructed usng much shorter consttuent block codes.

3 Performance Comparson of Product Codes and Cubc Product Codes usng FRBS for Robust Watermarkng 59 Ths concept can also be vewed as that product code Ω s ntersecton of two codes A and A. Where A s a code represented by all n2 n matrces whose each row s a member of code, smlarly s a code represented by all n n 2 A A 2 matrces who s each column s a member of code Ths can be wrtten as; Ω A A (2) 2 B. Cubc Product Codes In cubc block codes (CPC) all three dmensons are encoded by three dfferent lnear block codes. In ths paper, Bose Chaudhur Hocquenghem (BCH) codes [] are consdered as consttuent codes n the constructon of CPCs. Let there be three BCH codes, namely A, A 2 and A wth the parameters [ N, K, D ] and [ N, K, D ] N, K 2, A 2 [ N, K, D ] respectvely. and represent codeword length, message length and mnmum hammng dstance (d mn) of the code A, respectvely and =, 2,. The code rate of the consttuent codes n D product codes can be wrtten as; D K R,,2, () N The D product code can be constructed n the followng manner.. Place K K 2 K nformaton bts n a cube lke structure such that s heght, depth of the cube 2. Encode K Krows usng code A 2, whch wll result n K N 2 K szed cube. Encode N 2 Krows usng code A, whch wll result n N N 2 K szed cube 4. Encode N N 2rows usng code A, whch wll result n N N 2 N szed cube. Ths s the fnal codeword of the cubc product code. Ths process s shown n fg-2. The parameters of the resultant cubc product code are gven as [ N, K, D ], where N K N N N 2 K K K 2 D D D D 2 (4) K K 2 as wdth and K as And the resultant code rate of the CPC can be gven as; R R (5) Snce CPCs possess a hgh mnmum dstance compared to ther consttuent codes, they have a much better error correcton capablty also. The error correcton capablty t s gven as; dmn t floor[ ] (6) 2 Cubc product block codes can also be consdered as the Cartesan product of ts consttuent lnear block codes. Also n some defntons t s also consdered as the ntersecton of the consttuent lnear block codes. Hence two dfferent notons are used as gven n Equ-5 and Equ-6.. X A A2 A (7) where represents Kronecker product of two codes and X s resultant product code. Also t can be vewed as; X A A A (8) 2 Hence ths can also be wrtten as the cubc product code s ntersecton of three codes that are, where s a code represented by all A :,2, N N N 2 each element s a member of code represented by all N N N 2 element s a member of code element of s a member of code Fgure 2. A 2 A A cubc matrces whose. Smlarly A 2 s a code cubc matrces whose each and A. A Cubc Product Codes C. Modfed Iteratve Decodng Algorthm s a code whose each Modfed Iteratve Decodng Algorthm (MIDA) was proposed by [4] and t s a revsed verson of Iteratve decodng algorthm orgonally proposed by [5] whch s a suboptmum decoder. MIDA s a hard decson decoder.. It s actually a lst decoder n whch separate lsts are generated for each row n row code and each column n column code or the product code. Syndrome decodng of lnear block codes s used for complexty reducton [6]. In ths way number of rows/columns, for whch lsts are to be bult, s reduced sgnfcantly, hence the decodng complexty s reduced sgnfcantly. Also the complexty reducton grows sgnfcantly wth each passng teraton n decodng. MIDA has been appled for decodng varous codes used for dgtal mage watermarkng as well as for adaptve communcaton systems lke adaptve frequency dvson multplexng (AOFDM) systems n the lterature [7-20].

4 60 III. Human Vsual System All areas of the mage are not equally sutable for embeddng the watermark nformaton because dfferent regons may have dfferent level of senstvtes. For example, unform areas of mage are very senstve to the addton of watermark nformaton so, only small amount of nformaton can be added n the unform areas whereas, the edge areas can support for embeddng greater watermark nformaton. The Human vsual system (HVS) has been consdered wth several phenomenon that permts to adjust the pxel values to elude percepton [2]. The FRBS has been used here to adapt the HVS dfferent propertes. In ths scheme, we are consderng texture, brghtness and edge senstvty, so that embeddng the watermark nformaton n these features makes the mage mperceptble. A. Brghtness Senstvty As brghter background areas are less senstve than the dark ones, pxels wth hgh values of brghtness are chosen for embeddng the watermark bts. It s common practce to represent the mages n 8-bt format, that s one byte per pxel, each of whch has a value from 0 to 255. The pxel value 0 represents the maxmum darkness n the mage whle the value 255 represents the maxmum brghtness n the mage whereas, the grey shades represent the values n between these lmts. B. Edge Senstvty To make the lower vsblty of embedded sgnal, hgher edges are chosen for embeddng the watermark []. A gradent analyss has been made to test the model by usng dfferent edge detecton methods such as sobel, prewtt and canny. In the present scheme, we have used Canny method for calculatng edge senstvty. C. Texture Senstvty S E The stronger the texture features, the lower s the vsblty of the embedded data so, our scheme search for the pxels wth the hghest texture for embeddng the watermark data. Texture senstvty bascally measures the actvty of the center pxel wth ts neghbors (fg-), Rahman et al. the postulaton that the texture has locally two pared aspects, strength and the pattern. The effectveness has been proposed to be an operatve descrptor n texture classfcaton [5]. In expermental studes, LBP has became the strongest measure for texture analyss whch can be comprehended as a unversal methodology to the tradtonally dfferent statstcal and physcal models of texture analyss [6]. The most mportant property of LBP operator n real world applcatons s ts nvarance aganst monotonc gray level changes. LBP s defned as a gray-scale nvarant texture measure, resultng from a descrpton of texture n a local neghborhood. A bnary value from 0 to 255 s obtaned by concatenatng the values of the neghborhood results n a clock wse drecton for each pxel. In the present scheme, LBP method for calculatng texture senstvty s used. IV. Human Vsual System S T Here a Fuzzy Rule Based System s used to fnd those regons n the mage where more nformaton can be embedded. Ths decson s based the HVS factors dscussed n prevous secton. FRBS decdes that how much data can be embedded n the whch regons of the mage wth a sgnfcant level of mperceptblty. A. Desgn of Fuzzy Rule Based System As mentoned earler, frst FRBS has three nput varables namely brghtness senstvty, texture senstvty and edge senstvty duly defned n prevous secton. The nput range of brghtness and texture senstvty s between and edge senstvty could ether be 0 or. Fve membershp functons are used to cover the nput space of brghtness senstvty (very dark, dark, dm, brght and very brght), two membershp functons are used to represent edge senstvty (low, hgh) and fve membershp functon for texture senstvty (very smooth, smooth, average, rough and very rough). These relatonshps for the three nput varables are shown n fg-4, fg-5 and fg-6 respectvely. There s one output varable named capacty factor (alpha). Fve membershp functons (very low, low, medum, hgh and very hgh) are used to cover the range whch s between 0 and as shown n fg-7. As cardnalty the of rule base s the cartesan product of number of membershp functons n each nput varables, there are ffty rules n the rule base, so the rule base s complete as t contans rule for each nput combnaton. As all three features are somewhat n drectly proportonal to the output, the rules are formulated accordngly. The possble values of varable edge senstvty are 0 or, so twenty-fve rules are formulated for each case. Rules can be found n table- and table-2 for edge senstvty and 0 respectvely. Each table contans twenty-fve rules. The frst row and frst column of each table contans IF part whle rest of the table contans accordng value of THEN part. A rule format can be expressed as; Fgure. Neghbours of a pxel IF (Texture= Average AND Brghtness = Dm AND Edge = ) THEN (Alpha = Medum ) Actvty g g g g g g g g (9) c 0 c c 2 c Ojala et al. [4] proposed a local bnary pattern (LBP) operator for calculatng texture senstvty that was based on

5 Performance Comparson of Product Codes and Cubc Product Codes usng FRBS for Robust Watermarkng 6 Inference Engne: Mamdan Inference Engne (MIE) s used for nferrng that an nput vector s mapped on to whch correspondng/approprate output pont/value by makng use of the rules and ther weghts n the rule base. In MIE, fuzzy operaton AND s chosen as MIN whle OR s chosen as MAX. Fgure 4. Input varable Brghtness senstvty (Bs) De-Fuzzfer: Standard Center Average Defuzzfer (CAD) s used to transform the fuzzy output value nto the closest and accurate crsp value. CAD s preferred among ts peers due to ts effectveness as well as less computatonal requrements durng de-fuzzfcaton process. Fgure 5. Input varable Edge senstvty (Es) Fgure 7. Output varable The Capacty factor (alpha) Fgure 6. Input varable Texture senstvty (Ts) The surface vews of the rule base between texture and brghtness for edge- and edge-0 are shown n fg-8 and fg-9 respectvely. Both of these fgures narrate that hgher the values of brghtness and texture senstvty, mage capacty factor, alpha, s hgher and vse versa. However, ths mpact s even more when the edge senstvty s and less when edge senstvty s 0, whch conforms to the defntons gven n prevous secton. That s regons wth edge senstvty one, can carry more nformaton and vce versa. B. Components of Fuzzy Rule Base System Fuzzfer: Standard Gaussan fuzzfer s used to transform crsp values of nput nto correspondng fuzzy values. Among other fuzzfers, Gaussan has ts own sgnfcance due to contnues approxmatons unlke trangular and trapozodal fuzzfers, where mappng s not that fne. Texture Edge = Texture Brghtness V-Low Low Dm Hgh V-Hgh V-Smooth VL L L M M Smooth L L M M H Average L M M H H Rough M M H H VH V-Rough M H H VH VH Table. Rulebase wth Edge Senstvty= Edge = 0 V- Smooth V- Low Brghtness Low Dm Hgh VL VL L L M Smooth VL L L M M Average L L M M H Rough L M M H H V- Rough V- Hgh M M H H VH Table 2. Rulebase wth Edge Senstvty=0

6 62 Rahman et al. Fgure 8. Surface vew wth edge senstvty = Fgure 9. Surface vew wth edge senstvty = 0 Two dfferent BCH codes are used for the experment. Frst one s [6,,4], n whch the mnmum dstance s 4, that means the error correcton capablty s. The second one s [,2,5] wth a mnmum dstance of 5, that means t s a two bt error correcton code. Same codes are used for constructon of product codes as well as cubc product codes, so that comparson can be made more nterestng. Moreover, same code s used n both dmensons of PC and all three dmensons of CPC. Fg-0 shows the performance of Product Codes n terms of bt error rate (BER) over an Addtve Whte Gaussan Nose (AWGN) channel. It s apparent from the fgure, that the Product Code wth [,2,5] as component code performs better than Product Code wth [6,,4] as component code. Ths s because the former code has a greater mnmum dstance and greater error correcton capablty. Fg- shows the performance of CPC n terms of BER over an Addtve Whte Gaussan Nose (AWGN) channel. It s apparent from the fgure, that the CPC wth [,2,5] performs better than CPC wth [6,,4]. Ths s because ths code has a greater mnmum dstance and better error correcton capablty. One thng that s noteworthy, product codes wth same consttuent codes can acheve same performance as CPC but wth a greater sgnal to nose rato (SNR). For example, to acheve a BER of 0e-5, product code [,2,5] 2 demands almost 26dB SNR, whle same code n CPC demands almost db SNR. Smlarly, n the case of [6,,4] consttuent code, Product code demands almost 27dB SNR s requred to acheve BER 0e-4, whle n CPC we need almost dbs to acheve the same BER performance whch s agan half than the prevous case. So from the fgures t s clear that product codes need double SNR compared to CPC n order to acheve same BER performance. That s manly because the thrd dmenson s payng n terms of performance mporvement. V. Smulaton Results In ths secton, effectveness of the CPCs s shown ntally n terms of bt error rate (BER) and ther mmunty aganst certan attacks n dgtal mage watermarkng. The smulaton parameters used n the experment, are lsted n table- gven below. Sr. # Parameter Value Consttuent code- [6,,4] 2 Consttuent code-2 [,2,5] Product Code- [6,,4] 2 4 Product Code- [,2,5] 2 5 Cubc Product [6,,4] Code- 6 Cubc Product [,2,5] Code- 7 Decoder MIDA [] 8 Attack type- AWGN 9 Attack type-2 Salt & Pepper nose 0 Attack type- JPEG Compresson Table. Smulaton Parameters Fgure 0. Performance of Product codes over AWGN Though the decodng complexty of CPC s more that Product codes, however, ther BER performacne s sgnfcantly greater than that of product codes. To make ths comparson crystal clear, consttuent codes for both product codes and cubc product codes are kept smlar so that the comparson can made on the structures rather than consttuent codes.

7 Performance Comparson of Product Codes and Cubc Product Codes usng FRBS for Robust Watermarkng 6 the recovered watermark s shown after extractng from the host mage. The peak sgnal to nose rato (PSNR) whch s consdered as the perceptual measure of orgonal and watermarked mage. Ths s actually the measure of mperceptblty that how much orgonal and watermarked mages are dfferent. That can be wrtten as [0]: PSNR 0log 0 ( ) MSE where (0) M N MSE [ f ( x, y) f ( x, y)] MN 0 j 0 2 Fgure. Performance of CPC over AWGN Fg-2 shows the orgnal cover mage of Lena whle the watermark mage s shown n fg-. It s a custom desgned watermark, just to make t compatble (n all dmensons) for encodng wth the product codes and CPC beng used respectvely. Fgure 2. The cover mage Fgure. The watermark mage After encodng the mage by the Product Code wth [,2,5] (beng superor between the consttuent codes, t s the only code that s used n subsequent comparsons) as the consttuent code, the watermark s embedded n the regons of the orgnal mage selected by the Fuzzy Rule Based System (FRBS) dscussed n the prevous secton. The watermarked mage s shown n fg-4. From the naked eye vew, t s clear that the mage s stll mperceptble. Ths s due to the pxels selected by FRBS. The mage s then offered a number of attacks and Here M and N are dmensons of the mage, mean squared error (MSE) s a normalzed dfference between orgonal mage and the watermarked mage. Though the practcal level of mperceptblty vares from applcaton to applcaton, however, n the lterature 50dB or hgher level of PSNR s consdered as a good mperceptblty [0]. After embeddng the watermakr encoded by the product code the PSNR of the mage s found 70dB whch s practcally a very good level of mperceptblty. f ( x, y) f ( x, y) After encodng the mage by the CPC wth [,2,5] as the consttuent code, the watermark s embedded n the regons of the orgnal mage selected by the Fuzzy Rule Based System (FRBS) dscussed n the prevous secton. The watermarked mage s shown n fg-5. From the naked eye vew, t s clear that the mage s stll mperceptble. Ths s due to the pxels selected by FRBS. The mage s then offered a number of attacks and the recovered watermark s shown after extractng from the host mage. After embeddng the watermakr encoded by the product code the PSNR of the mage s found 6dB whch s practcally stll a good level of mperceptblty. The dfference n PSNR n fg-4 and fg-5 s because n case of CPC, we have to embed more nformaton n terms of watermark snce our watermark s encoded n three dmenson whch results n relatvely bgger nformaton contents. Ths would help us n enhanced robustness but at the cost of lttle reduced mperceptblty. Ths s depcted n subsequent fgures. Fg-6 shows the recovered watermark after speckle nose attack wth varance 0.0 resultng n Nc=0.796 n case of Product Codes whle Nc = Here Nc s the smlarty ndex between orgonal mage and the recovered mage. So hgher the ndex, more robust wll be the mage. Though requred value of Nc vares from applcaton to applcaton however, as a common practce, value of Nc hgher than 0.7 s consdered as a good level of robustness [0]. From fg-6 t s aparant that CPCs are more robust compared to the smple product codes. Fg-7 shows the recovered watermark after the attack of Gaussan nose wth varance 0.0, showng Nc = 0.7 n case of Product codes whle Nc=0.84 n case of the CPC. Fg-8 shows the recovered watermark after attack of Salt & pepper nose wth varance 0.0, havng Nc = 0.86 n case of CPC and Nc=0.80 n case of Product codes.

8 64 Fg-9 shows the recoverd watermark after rotaton attack of 2 degrees, havng Nc = n the case of product codes whle Nc=0.95 n case of CPC. From fg-6 to fg-9, t can easly be deduced that proposed schemes are sgnfcantly robust aganst the sad attacks. Moreover, Cubc product codes outperforms than Product codes n terms of robustness. The proposed schemes are compared wth each other for dfferent knds of wellknown attacks found n the lterature and the comparson s enlsted n table 4. From both columns t s clear that CPC performs better than Product codes n terms of robustness. Rahman et al. Fgure 6. Recoverd watermark after attack of speckle nose wth varance 0.0 showng (a) Nc = case of product code (b) Nc = 0.87 case of cubc product code Attack Type Product Codes (Nc) Cubc Product Codes (Nc) Speckle Nose Gausson Nose Salt & pepper Nose Rotaton of Degree Table 4. Comparson of schemes Fgure 7. Recoverd watermark after attack of Gaussan nose wth varance 0.0, havng (a) Nc = 0.7 case of product codes (b) Nc = 0.84 case of cubc product codes Fgure 4. The watermarked mage (havng PSNR=70dB) Fgure 8. Recoverd watermark after attack of Salt & pepper nose wth varance 0.0, havng (a) Nc = 0.80 case of product codes (b) Nc = 0.86 case of cubc product codes Fgure 5. The watermarked mage (havng PSNR=6dB) Fgure 9. Recoverd watermark after rotaton attack of 2 degrees, havng (a) Nc = case of product codes (b) Nc = 0.95 case of cubc product codes

9 Performance Comparson of Product Codes and Cubc Product Codes usng FRBS for Robust Watermarkng 65 VI. Concluson Ths paper presents a novel dea of usng product codes and cubc product codes (CPC) for makng the watermark robust n a dgtal mage watermarkng scenaro. Product codes are two dmensonal lnear block codes whle CPC are three dmensonal lnear block codes. The structure of both types of codes make them compatble and sutable wth the dgtal mage watermakng. In a number of applcatons, the watermark s more mportant as compared to the cover mage, so makng the watermark robust s the major focus of the feld. Lke when the cover mage s publcally avalable and accessble but ts ownershp s stll needed to be preserved. A fuzzy rule based systems (FRBS) s proposed to hghlght the areas of the cover mage (pxels) where the watermark can be nserted wth a sgnfcant level of mperceptblty. Ths s done by fuzzfyng and adequately utlzng the parameters of Human Vsual System (HVS). Due to ther error correctng capabltes, product codes and CPCs has shown the notable performance n the results n terms of robustness aganst varous attacks. Moreover, t was observed that CPCs performs better than Product codes n terms of robustness whle usng Product codes may result n a lttle degradaton n mperceptblty of the mage that s stll somewhat neglgble. Comparsons are made through computer smulatons usng MATLAB Acknowledgment Ths research was partally funded by Baran Insttute of Informaton Technology (BIIT), Rawalpnd, Pakstan. References [] Gokozan, T., Template Based Image Watermarkng n the Fractonal Fourer Doman, MSc thess, Mddle East Techncal Unversty (METU), [2] P. Elas, Error-free codng, IEEE transactons on Informaton Theory, vol. 4, pp. 29-7, 954. [] Atta-ur-Rahman, Quresh I.M., Effectveness of Modfed Iteratve Decodng Algorthm for Cubc Product Codes, proceedngs of Internatonal Conference on Hybrd Intellgent Systems (HIS 4), pp , December 204. [4] C. Nanjunda, M. Haleem and R. Chandramoul, Robust encrypton for secure mage transmsson over wreless channels, [5] C. Mathur, K. Narayan and K. Subbalakshm, On the desgn of error-correctng cphers, EURASIP Journal onwreless Communcatons and Networkng, Vol.2006, pp.-2, November [6] M. A. El-Iskandaran, S. Darwsh, S. M. Abuguba, Relable Wreless Error Correcton Technque for Secure Image Transmsson [7] A. Gabay, M. Keffer, P. Duhamel, Jont Source-Channel Codng Usng Real BCH Codes for Robust Image Transmsson. IEEE transactons on Image Processng. Vol. 6, No. 6, June [8] J. M. Zan and A. R. M. Fauz, "Medcal Image Watermarkng wth Tamper Detecton and Recovery," n Proceedngs of 28th IEEE EMBS Annual Internatonal Conference, pp , [9] Atta-ur-Rahman, Naseem, M.T.; Quresh, I.M.; Muzaffar, M.Z.; "Reversble watermarkng usng Resdue Number System". IEEE 7 th Internatonal Conference on Informaton Assurance and Securty (IAS ), pp , 5-8 December 20, Melacca, Malaysa. [0] M.T.Naseem, I.M.Quresh, Atta-ur-Rahman, Robust Watermarkng for Medcal Images Resstant to Geometrc Attacks, IEEE 5 th Internatonal Mult-topc Conference (INMIC 5), pp , December 205, Islamabad. [] Atta-ur-rahman, Quresh I. M. and Naseem M.T., A Novel Technque for Relable Image Transmsson usng Product Codes. Internatonal Journal of Computer Applcatons (IJCA) Vol. 65(4), pp. 2-7, March 20. [2] Atta-ur-Rahman, N. Saba, N. An, A Novel Robust Watermarkng Scheme usng Cubc Product Codes, Journal of Informaton Securty Research, vol. 6(), pp. 4-24, 205. [] J. Canny, A computatonal approach to edge detecton, IEEE Transactons on Pattern Analyss and Machne Intellgence, no. 6, pp , 986. [4] Atta-ur-Rahman, S.A.Ghaur, H.Adeel, A.Waheed, Performance of Modfed Iteratve Decodng Algorthm for Product Codes, Internatonal Conference on Computatonal Aspects of Socal Networks (CaSoN ), pp [5] Al-Askary, O., Codng and teratve decodng of concatenated mult-level codes for the Raylegh fadng channel, Doctoral thess n Rado communcaton systems, Stockholm, Sweden, [6] Davd G. Forney Jr. Generalzed Mnmum Dstance Decodng. IEEE Transactons on Informaton Theory, vol. 2, no. 2, pp.25-, 966. [7] Atta-ur-Rahman, I.M.Quresh, M.T.Naseem, M.Z.Muzaffar, A GA-FRBS based Rate Enhancement Scheme for OFDM based Hyperlans, Fronter n Informaton Technology (FIT 2), pp. 5-58, Pakstan. [8] Atta-ur-Rahman, I.M.Quresh, Effectveness of Modfed Iteratve Decodng Algorthm for Cubc Product Codes, 4 th Internatonal Conference on Hybrd Intellgent Systems (HIS 4), pp , 4-6 December, 204. [9] Atta-ur-Rahman, I.M.Quresh, M.T.Naseem, M.Z.Muzaffar, A Fuzzy Rule Base Aded rate enhancement Scheme for OFDM Systems, Internatonal Coference on Emergng Technologes (ICET 2), pp. -6, 8-9 October, Pakstan. [20] Atta-ur-Rahman, M.T.Naseem, M.Z.Muzaffar, Reversble and Robust Watermarkng usng Resdue Number System and Product Codes, Journal of Informaton Assurance and Securty (JIAS), Vol. 7, pp. 56-6, 202. [2] T. Ojala, Petkänen, M. and Mäenpää, T. Multresoluton gray-scale and rotaton nvarant texture classfcaton wth local bnary patterns, IEEE Trans. PAMI, vol. 24, pp , [22] R. M. Haralck, Shanmugam, K. and Dnsten, I. H., Textural features for mage classfcaton, IEEE Transactons on Systems, Man and Cybernetcs, no. 6, pp , 97. [2] M. Petkänen, Image analyss wth local bnary patterns. Image Analyss, Sprnger Berln Hedelberg, pp. 5-8, Author Bographes Dr. Atta-ur-Rahman s currently workng at Baran Insttute of Informaton Technology (BIIT), Rawalpnd, Pakstan, as Assocate Professor & Deputy Drector (R&D). He has completed hs BS degree n Computer Scence from Unversty of The Punjab, Lahore, Pakstan; MS degree n Electronc Engneerng from Internatonal Islamc Unversty, Islamabad, Pakstan and PhD degree n Electronc Engneerng from ISRA Unversty, Islamabad Campus, Islamabad, Pakstan n years 2004, 2008 and 202, respectvely. Hs research nterests nclude nformaton and codng theory, wreless/dgtal communcaton, dgtal sgnal processng, dgtal mage watermarkng, soft/evolutonary computng and hybrd ntellgent systems (HIS).

10 66 Rahman et al. Muhammad Azam s currently workng wth Medcal Transcrpton and Bllng Company (MTBC), Rawalpnd, Pakstan, as software qualty assurance (SQA) Engneer. He has completed hs BS degree n Computer Scence from PMAS-AA Unversty Rawalpnd, Pakstan n year 20. He s currently pursung hs MS degree n Computer Scence from Baran Insttute of Informaton Technology (BIIT), PMAS-AA Unversty, Rawalpnd, Pakstan. Hs research nterests nclude applcatons for AI and Data Mnng algorthms for ntellgent SQA. Gohar Zaman s currently workng wth The Unversty of Lahore as Lecturer n Computer Scence. He has completed hs MSc n Computer Scence from Natonal Unversty of Modern Languages (NUML), Islamabad, Pakstan and MS n Computer Scence from The Unversty of Lahore, Islamabad Campus, Pakstan n years 2009 and 20 respectvely. He s currently pursung hs PhD n Computer Scence from Shaheed Zulfqar Al Bhutto Insttute of Scence and Technology (SZABIST), Islamabad Campus, Pakstan. Hs research nterests nclude AI, data and text mnng, automatc text classfcaton/categorzaton (ATC) and fuzzy systems.