IMAGE AUTHENTICATION AND RECOVERY USING WAVELET-BASED DUAL WATERMARKING

Transcription

1 U.P.B. Sc. Bull., Seres C, Vol. 77, Iss. 4, 015 ISSN IMAGE AUTHENTICATION AND RECOVERY USING WAVELET-BASED DUAL WATERMARKING Radu Ovdu PREDA 1, Ioana MARCU, Amela CIOBANU 3 In ths paper a novel atermarkng scheme for mage authentcaton and recovery s presented. The algorthm can detect modfed regons n mages and s able to recover a good approxmaton of the orgnal content of the tampered regons. For ths purpose, to dfferent atermarks have been used: a sem-fragle atermark for mage authentcaton and a robust atermark for mage recovery, both embedded n the Dscrete Wavelet Transform doman. The proposed method acheves good mage qualty th mean Peak Sgnal-to-Nose Rato values of the atermarked mages of 4 db and dentfes mage tamperng of up to 0% of the orgnal mage. Keyords: multmeda securty, mage forgery detecton, dgtal atermarkng, mage authentcaton, mage recovery, Dscrete Wavelet Transform 1. Introducton Dgtal content, such as mages, vdeo and audo can be easly coped and dstrbuted through dfferent communcaton channels. The avalablty of poerful sgnal processng tools makes t very dffcult to guarantee the ntegrty of multmeda content. Dgtal mages are used n legal dsputes nvolvng tampered pctures publshed n nespapers and magaznes, accdents, poltcal or celebrty scandals, etc. Under these crcumstances, n order to prevent malcous, ntentonal tamperng, mage authentcaton has become a very mportant and challengng ssue n the dgtal orld. One of the best solutons for mage authentcaton s dgtal atermarkng, a process by hch a user specfed sgnal (atermark) s hdden or embedded n the orgnal mage. A great number of scentfc publcatons n ths feld only authentcate the content of dgtal mages and are not able to reconstruct the orgnal content [1-6]. Most technques use fragle atermarks for authentcaton [3-7]. These atermarks can detect malcous alterng of the mage content, but are also destroyed even after the smallest unntentonal modfcaton, an undesred property n most applcatons. Compared to these methods, the technque 1 Assoc. Prof., Telecommuncatons Dept., Unversty POLITEHNICA of Bucharest, Romana, e- mal: radu@comm.pub.ro Lect., Telecommuncatons Dept., Unversty POLITEHNICA of Bucharest, Romana 3 Lect., Telecommuncatons Dept., Unversty POLITEHNICA of Bucharest, Romana

2 00 Radu Ovdu Preda, Ioana Marcu, Amela Cobanu proposed n ths paper uses a sem-fragle authentcaton atermark, hch s able to thstand a good degree of common mage processng. Many other sem-fragle authentcaton technques use block-based approaches n the spatal or Dscrete Cosne Transform (DCT) doman to detect the tampered regons [6, 8, 9]. Most of these schemes are vulnerable to counterfetng attacks, lke the Vector Quantzaton (VQ), and the tamper detecton resoluton s lmted to the block sze. Smaller block szes and hgher atermark payloads are necessary to mprove the detecton resoluton, resultng n a consderable degradaton of the mage qualty. To allevate these problems, the technque proposed n ths paper uses Wavelet coeffcents permuted th a random key to embed the authentcaton atermark, protectng the scheme aganst local attacks. The embeddng of the authentcaton atermark n the Wavelet doman also assures a better detecton resoluton than block-based methods. Another desred property of an authentcaton scheme s the ablty to recover the orgnal content of the detected tampered regons. Only a small percentage of the exstng algorthms are able to do ths, because t comes th a trade-off: the use of a second atermark, the recovery atermark, an approxmaton of the orgnal mage th hgh payload, results n further degradaton of the mage qualty [10-1]. Dfferent recovery schemes try to reduce the payload of ths atermark usng compresson. In [13], Chamla et al. use a hghly compressed verson of the orgnal mage as a recovery atermark, obtaned by applyng the DCT to the second level Wavelet approxmaton subband, and embed ths atermark n some mddle frequency Wavelet coeffcents. Ths approach s fragle to any knd of mage processng operaton and also to large content alterng modfcatons and s not able to recover the dgest mage. A recovery scheme th better results s proposed n [14], here the dgest mage s compressed usng arthmetc codng and protected by applyng a Bose-Chaudhur- Hocquenghem (BCH) error correctng code. Ths atermark s embedded n the mddle sub-band detal Wavelet coeffcents usng a Least Sgnfcant Bt (LSB) approach. The error correcton code ncreases the atermark payload and degrades the mage qualty and, even f t s able to correct some errors produced by salt&pepper nose n the extracted arthmetc code, the LSB method s qute fragle to any modfcatons. Most of the exstng recovery schemes use ether a fragle recovery atermark (hghly compressed verson of the orgnal mage) [13], [15, 16], or a fragle embeddng strategy, lke LSB embeddng [4], [9], [14], [17]. Because the recovery strategy must be as robust as possble to any knd of modfcaton of the mage, these methods do not produce satsfactory results. The recovery algorthm proposed n ths paper acheves not only good robustness to common sgnal processng operatons, but also to large malcous tamperng of the mage.

3 Image authentcaton and recovery usng avelet-based dual atermarkng 01 The paper s organzed as follos: n Sectons and 3 the proposed mage authentcaton and recovery scheme s presented, ncludng the block dagrams of the atermark embedder and decoder and the detaled steps of the algorthm. Secton 4 contans the expermental results and performances of the proposed scheme n terms of mage qualty, detecton, localzaton and recovery capablty and robustness to common mage processng operatons. Conclusons are gven n Secton 5.. The proposed dual atermark encodng scheme The block dagram of the embeddng scheme, gven n Fgure 1, conssts of to man blocks, for generatng and embeddng the to dfferent atermarks, the authentcaton and recovery atermark. These procedures are descrbed n the follong..1. Generaton and embeddng of the authentcaton atermark The authentcaton atermark generaton and embeddng procedure s presented n the loer half on Fgure 1 and s descrbed n the follong: A to level b-dmensonal Dscrete Wavelet Transform (D-DWT) s used to transform the orgnal, grayscale mage n the Wavelet doman. The detal Wavelet coeffcents of the LH, HL and HH Wavelet sub-bands of the second Wavelet decomposton are selected for embeddng the authentcaton atermark. By selectng hgher resoluton sub-bands for atermark embeddng the algorthm acheves a better localzaton of ntentonal tamperng. LL LH HL HH HL 1 LH 1 HH 1 Generaton and embeddng of the recovery atermark LL LH 1 HL 1 HH 1 Select n sgnfcant bts Key based permutaton Key based permutaton Redundant embeddng usng QIM LH C C m Key based Coeffcent Compute Modfy HL permutaton groupng mean mean HH g Authentcaton K 1 K 1 atermark Generaton and embeddng of the authentcaton atermark K d 1 Fg. 1. The proposed atermark embeddng scheme d at m K Inverse permutaton Inverse permutaton K 1 IDWT Watermarked mage

4 0 Radu Ovdu Preda, Ioana Marcu, Amela Cobanu Vector C contanng the Wavelet coeffcents of the three sub-bands s randomly permuted th the use of a secret key K 1 nto a vector C. Ths permutaton ensures the separaton of coeffcents from the same spatal locaton. C Is dvded nto groups of g coeffcents, here g controls the atermark capacty of the scheme. A bt of the authentcaton atermark ll be embedded n every group of g coeffcents. The use of a smaller group sze ll have a bgger atermark payload and thus a hgher degradaton of the mage qualty as an effect, but, on the other hand, t ll not decrease the detecton resoluton. The authentcaton atermark s a bnary random sequence, generated based on a secret key K 1 and has the same length as the number of coeffcent groups. The eghted mean m of a group of permuted avelet coeffcents s calculated accordng to (1): d j m = ( 1) c( j) (1) j= 1 here c ( j ) s the jth coeffcent of group I and ( 1) j s the eghtng factor used to make the scheme more reslent aganst common mage processng. Such unntentonal alteratons usually change the entre mage content and do not modfy the eghted mean. To embed a atermark bt n a group of coeffcents, the eghted mean s quantzed to the nearest even or odd quantzaton level accordng to the value of the correspondng atermark bt, usng (). mod ( ) m Q Q f m Q = m =, () m Q Q+ Q f mod ( m Q ) here m s the atermarked mean, mod s the remander after dvson by and. s the nteger part operator. The eghted mean m of every group of coeffcents s changed to the atermarked mean m by modfyng the Wavelet coeffcent c,max ( j ) of the hghest magntude. The random permutaton ensures that every group has at least one coeffcent th hgh magntude. Coeffcent c ( ),max j s modfed usng (3): j c,max ( j) = c,max ( j) + ( 1) sgn( c,max( j)) ( m m), (3) th c ( ),max j beng the atermarked coeffcent and m

5 Image authentcaton and recovery usng avelet-based dual atermarkng 03 1, f x 0 sgn( x) = (4) 1, f x > 0 The Wavelet coeffcents are shuffled back to ther orgnal poston usng the nverse permutaton th the same secret key K 1, obtanng the ne atermarked sub-bands LH, HL and HH... Generaton and embeddng of the recovery atermark The generaton and embeddng of the recovery atermark are performed n the upper part of Fgure 1. The follong steps are performed: The LL approxmaton sub-band, s used as the recovery atermark, also called the dgest mage. It s a reduced verson of the orgnal mage. To further reduce the atermark payload, only the frst n most sgnfcant bts of every avelet coeffcent from LL are used for embeddng. Let 1 denote the bnary atermark vector of length l at, gven n (5), here M N s the resoluton of the mage. lat = nmn /16 (5) To ncrease the securty of the algorthm the vector 1 s shuffled usng a random permutaton based on the secret key K, obtanng the permuted bnary atermark. The detal Wavelet coeffcents of the LH1, HL1and HH1 sub-bands are used for embeddng the recovery atermark. For securty reasons, they are shuffled usng a thrd key K 3, obtanng the vector d of shuffled coeffcents of sze l coef : lcoef = 3 MN /4 (6) Each bt of the recovery atermark s redundantly embedded nto every coeffcent of a group of h coeffcents of vector d, here h s obtaned usng (7): h = lcoef / lat = 1/ n (7) A atermark bt s embedded nto a Wavelet coeffcent usng a Quantzaton Index Modulaton approach, as shon n (8), here d at s the vector of atermarked coeffcents. dj ( ) Q dat ( j) = Q+ ( ), = 1, lat, j = 1, h Q (8) After the entre atermark has been embedded, an nverse permutaton of the atermarked coeffcents s done, ensurng that a atermark bt s spread n the entre mage. Let LH1, HL 1 and HH 1 denote the avelet sub-bands contanng the recovery atermark.

6 04 Radu Ovdu Preda, Ioana Marcu, Amela Cobanu To obtan the atermarked mage, the Inverse D-IDWT s appled to tmes, frst on the coeffcents of the, LL LH, HL and HH sub-bands, obtanng the approxmaton sub-band LL 1, and the second tme on the coeffcents of the LL1, LH1, HL 1 and HH 1 sub-bands. 3. The proposed atermark retreval scheme A block dagram of the atermark decoder s gven n Fgure. The upper part of the fgure contans the mage recovery system and the bottom part contans the blocks to authentcate the mage and localze possble ntentonal tamperng. 3.. Retreval of the authentcaton atermark and mage authentcaton The extracton process of the authentcaton atermark and the steps for mage authentcaton are descrbed n the follong: Frst, the test mage undergoes a D-DWT decomposton. The LH, HL and HH Wavelet sub-bands are used to extract the authentcaton atermark. Wth the use of the secret key K 1, the same random permutaton from the encoder sde s performed on the vector of atermarked avelet coeffcents. The eghted mean m of every group of g coeffcents s calculated usng (1). Retreval of the recovery atermark LL LH HL HH LH 1 LH 1 HL 1 HH 1 Key based permutaton Watermark extracton usng QIM Decson functon Inverse permutaton Zero paddng HL 1 HH 1 K Recovered mage Blnear nterpolaton Medan flterng LL K 1 g LH HL HH Key based permutaton Coeffcent groupng Compute mean Watermark extracton Extracted atermark K 1 Authentcaton and tamperng localzaton Image authentcaton Morphology operatons Flterng Inverse permutaton Possbly tampered groups Watermark comparson Fg.. The proposed atermark decoder and mage authentcaton scheme Locally generated atermark

7 Image authentcaton and recovery usng avelet-based dual atermarkng 05 From every mean a atermark bt s extracted usng (9). m = round mod Q (9) Usng the secret key K 1, the orgnal atermark s locally generated and compared to the extracted one. If they match, the mage s declared as authentc. If not, the follong steps ll determne the authentcty and the locaton of possble tamperng. If a bt of the extracted atermark does not match the orgnal one, all coeffcents of group are consdered as potentally tampered. After permutng all coeffcents back to ther orgnal poston usng the secret key K 1, the potentally tampered coeffcents should be spread all over the second level detal Wavelet sub-bands. A hgh densty of potentally tampered coeffcents should ndcate that the correspondng regon has been tampered th. Authentc regons, on the other hand, should only contan solated flagged coeffcents, dstrbuted lke random nose. These coeffcents are false postves and should be consdered as authentc. As a result, the sub-bands LH, HL and HH ll contan flagged and nonflagged coeffcents. Let A be the bnary authentcaton matrx of sze ( M /) ( N /), the same sze as any of the three sub-bands. If there s a potentally tampered coeffcent at poston ( x, y ) n any of the three sub-bands, A( x, y) ll be set to 1. To remove the solated 1 bts from A, the authentcaton matrx ll be fltered usng both a nose removal flter and successve mathematcal morphology operatons th a dsk of a radus of one pxel as the structural element. The authentcaton matrx should no contan only regons of clustered flagged postons and should correctly ndcate the tampered locatons. To locate the regons that have actually been tampered th, the flagged postons n matrx A are mapped back to the spatal doman. Every poston n authentcaton matrx A of sze ( M /) ( N /) ndcates an actual regon of sze 4 4pxels n the mage, hch s the maxmum detecton resoluton of ths scheme. The quantzaton step sze Q, the flter sze and the sze of the structural element used for the arthmetc morphology operatons can be modfed to mprove the senstvty of the tamperng detecton.

8 06 Radu Ovdu Preda, Ioana Marcu, Amela Cobanu 3.. Retreval of the recovery atermark Frst, the three detal Wavelet sub-bands LH 1, HL 1 and HH 1 of the frst Wavelet decomposton are selected. A random permutaton th key K 3 of these coeffcents s performed on the selected Wavelet coeffcents, obtanng a vector d of shuffled coeffcents. From every coeffcent of vector d, a atermark bt s extracted accordng to (10), here h() s the extracted bt. d ( ) h( ) = round mod, = 1, lcoef Q (10) The vector h s dvded nto groups of h coeffcents. The atermark bt ( j) correspondng to the group j s obtaned by majorty votng, as n (11): hj h 0, f h( k) k= h( j 1) + 1 ( j) =, j = 1, lat. (11) hj h 1, f h( k) > k= h( j 1) + 1 The nverse permutaton th key K of vector s performed, obtanng. The bnary sequence s dvded nto groups of n bts and every group s padded th 8 n zeroes. Every group of 8 bts represents an extracted Wavelet coeffcent of the dgest mage (LL sub-band). After reorderng the coeffcents, the recovered mage LL of resoluton M N /16s obtaned. To remove the error pxels that could appear n LL because of ntentonal tamperng of the mage, a medan flter of sze 3x3 can be appled. Fnally, the recovered dgest mage s obtaned by blnear nterpolaton of the mproved verson of LL. The securty of the authentcaton and recovery system s ensured by three secret keys used to control the generaton of the authentcaton atermark and the random permutatons of Wavelet coeffcents. There can be dstnct keys for these three operatons or a sngle key. An attacker has to kno the secret key(s) n order to generate the correct coeffcent permutaton or a duplcate of the orgnal authentcaton atermark. Unlke most block-based methods, here the authentcaton of a block depends only on the content of the block tself, the proposed scheme s rendered mmune to the VQ attack by selectng the coeffcents of a group randomly from all over the Wavelet sub-bands.

9 Image authentcaton and recovery usng avelet-based dual atermarkng Expermental results We have used 100 mages of resoluton pxels to test the algorthm n terms of qualty of the atermarked mages, detecton and localzaton capablty of the tamperng, qualty of the recovered mages and robustness to common mage processng operatons. In Table 1 the mage qualty and decodng results of the proposed method are shon for dfferent quantzaton step szes Q and dfferent group szes g, here PSNR s the mean Peak Sgnal to Nose Rato for 100 test mages and BER s the mean decodng Bt Error Rate of the extracted authentcaton atermark, compared to the orgnal one. The borthogonal 4.4 Wavelet famly has been used, but t can be replaced th any other avelet famly, th mnmal mpact on the resultng mage qualty and detecton results. The PSNR has good values, above 40 db, except for Q=1, here the mean PSNR drops to 37 db. For every combnaton of parameters the authentcaton atermark can be extracted successfully thout any errors. Next, e have tested the capacty of the proposed approach to detect ntentonal tamperng of the mage and recover the tampered regons. For ths purpose, e have replaced a regon of dfferent szes of the atermarked mage th a regon of the same sze from another mage. Ths as done for every mage n the database. The szes of the tampered blocks ere 16 16, 3 3, 64 64, and19 19 pxels. After the tamperng, the authentcaton atermark s extracted from the tampered mage and the authentcaton algorthm returns the postons of the tampered regons. One of the major contrbutons of ths proposed technque s the capablty to repar the tampered regons. When a tampered regon s detected, the recovery algorthm ll recover the content of ths regon. These unauthentc parts of the mage are replaced th the correspondng regons of the recovered dgest mage. Mean PSNR and BER values Q d PSNR BER 4 46,99 0% ,51 0% 1 47,70 0% 4 40,95 0% ,47 0% 1 41,66 0% 4 37,41 0% ,9 0% 1 38,11 0% Table 1

10 08 Radu Ovdu Preda, Ioana Marcu, Amela Cobanu To measure the mage qualty of the reconstructed mages, e have calculated the mean PSNR values of these mages compared to the orgnal ones. The results are shon n Table, here all PSNR values are mean values for 100 mages. Table also contans the false negatve rate (FNR), the percentage of tampered mages declared as authentc. We can see, that for small tamperng and some embeddng parameters, lke g=4 and Q=1 e get a small percentage of false negatves, but for most parameters, ths percentage s zero. Fg. 3 shos to examples of mage authentcaton and restoraton. The frst mage s the popular Lena grayscale mage of sze51 51 pxels and the second mage s another test mage of resoluton pxels. Fg. 3a shos the atermarked mages usng a quantzaton step sze of Q=8 and a group sze of g=4. After embeddng the atermarked, the Lena mage has been modfed by replacng a regon of pxels th another regon of the same sze from the same mage (Fg. 3b, left mage). The second mage has been modfed by replacng the face of the oman n the top left regon th another oman s face and by removng the beer can and extendng the background (Fg. 3b, rght mage). In Fg. 3c the results of the authentcaton process are presented, here the hte regons denote the detected unauthentc part of the mages. The reconstructed mages are shon n Fg. 3d. We can see that the proposed algorthm s senstve to malcous manpulatons. It s able to correctly detect the tampered regon of the mage and to recover the mssng nformaton th acceptable qualty. Table Mean PSNR values of the recovered mages and false negatve rate for dfferent tamperng Sze of the tampered regon Q g PSNR FNR PSNR FNR PSNR FNR PSNR FNR PSNR FNR (db) (%) (db) (%) (db) (%) (db) (%) (db) (%) 4 44,18 0,00 41,79 0,00 38,74 0,00 33,7 0,00 9,31 0, ,31 0,00 4,1 0,00 38,6 0,00 31,49 0,00 3,13 0, ,08 0,00 41,9 0,00 38,56 0,00 3,49 0,00 3,13 0, ,73 10,00 38,57 3,33 37,04 0,00 3,77 0,00 9,03 0, ,8 0,00 39,0 0,00 37,7 0,00 3,04 0,00 3,1 0, ,45 0,00 39,18 0,00 37,08 0,00 6,89 0,00 3,14 0, ,55 13,33 35,98 3,33 34,97 0,00 31,45 0,00 8,30 0, ,6 10,00 36,51 0,00 34,84 0,00 31,6 0,00 3,74 0, ,45 10,00 36,61 0,00 35,13 0,00 7,87 0,00 3,19 0,00

11 Image authentcaton and recovery usng avelet-based dual atermarkng 09 (a) (b) (c) (d) Fg. 3. (a) atermarked mages for g=4 and Q=8, (b) tampered mages, (b) authentcated mages, (d) reconstructed mages Next e have tested the robustness of the proposed technque to mld mage processng operatons. The authentcaton atermark s a sem-fragle

12 10 Radu Ovdu Preda, Ioana Marcu, Amela Cobanu atermark, hch should be robust to common sgnal processng operatons that preserve the mage content, hle stll beng able to detect content alterng modfcaton n the mage. To demonstrate ths property, to scenaros have been used. Frst, the orgnal mages have been atermarked usng the best choce for the embeddng parameters (g=4, Q=8) and ere modfed usng the follong mage processng operatons: brghtenng/darkenng (lumnance of 40); addton of Gaussan nose of mean 0 and varance ; addton of salt&pepper nose, here 0.1% of the mage pxels ere modfed; JPEG compresson th a qualty factor of 85. Then, the modfed mages have been authentcated and the false postve rate (FPR) has been expermentally determned for the database of 100 test mages. The FPR s the percentage of authentc mages that have been declared as unauthentc. Table 3 shos the results for g=4 and Q=8, here all PSNR values n the table are mean values of the atermarked mages after applyng the mage processng operatons. For the second scenaro, the mages have been atermarked, modfed by replacng a regon of 64x64 pxels and fnally, an mage processng operaton has been appled. Then, the mages have been authentcated, the recovery atermark has been extracted and the orgnal content of the unauthentc regons has been reconstructed. The PSNR values from Table 3 are mean values for the reconstructed mages, compared to the orgnal ones. We have also expermentally determned the false negatve rates, hch are also gven n Table. The algorthm as able to successfully detect every content alterng modfcaton (FPR=0), but a small percentage of authentc mages, that have been affected by Gaussan nose, salt&pepper nose and JPEG compresson have been falsely declared as unauthentc. The proposed method has also been compared to to other approaches. Table 4 gves a summary of ths comparson. The man advantages of ths method, compared to the other technques, are the good mage qualty, the good detecton resoluton of 4x4 pxels for a small payload of the authentcaton atermark and the robustness to common mage processng operatons. Table 3 Robustness of the proposed scheme to common mage processng operatons Brghtenng Gaussan nose Salt&pepper nose JPEG Compresson No tamp. 64x64 No tamp. 64x64 No tamp. 64x64 No tamp. 64x64 PSNR (db) FPR PSNR FNR (%) (db) (%) PSNR (db) FPR (%) PSNR (db) FNR (%) PSNR (db) FPR (%) PSNR (db) FNR (%) PSNR (db) FPR (%) PSNR (db) 40,95 0,00 1,6 0,00 40,95 6,67 30,86 0,00 40, ,5 0,00 40,95 6,67 4,45 0,00 FNR (%)

13 Image authentcaton and recovery usng avelet-based dual atermarkng 11 Technque Image qualty for average parameters Comparson to other technques Detecton resoluton Robustness to common mage processng Table 4 Image recovery possble after removng [13] 36,65 db 4x4 Yes (only authentcaton atermark) Up to 5% of orgnal [14] 39,88 db 8x8 Only to salt&pepper nose Up to 10% of orgnal Proposed 39,73 db 4x4 Yes (both authentcaton and recovery atermarks) Up to 0% of orgnal 5. Conclusons In ths paper a novel Wavelet-based mage authentcaton and recovery scheme usng to atermarks has been proposed. The algorthm s blnd, semfragle, s able to detect and locate malcous tamperng n dgtal mages and recover a good estmate of the orgnal content even f the atermarked mage has been tampered th to a degree of 0%. The proposed method acheves hgh tamperng detecton resoluton and hgh mage qualty compared to other state of the art technques. The atermarkng scheme s protected aganst local attacks, lke the Vector Quantzaton attack, by randomzng the poston of Wavelet coeffcents used for embeddng th the use of a prvate key. The embedded authentcaton and recovery atermarks are also reslent aganst mld common mage processng operatons, lke brghtenng, addton of Gaussan and salt&pepper nose, and JPEG compresson, hle stll beng able to detect ntentonal tamperng th good accuracy. Acknoledgement Ths ork has been funded by the Sectoral Operatonal Programme Human Resources Development of the Mnstry of European Funds through the Fnancal Agreements POSDRU/159/1.5/S/ and POSDRU/159/1.5/S/ R E F E R E N C E S [1] H. M. Al-Otum, Sem-fragle atermarkng for grayscale mage authentcaton and tamper detecton based on an adjusted expanded-bt multscale quantzaton-based technque, Journal of Vsual Communcaton and Image Representaton, vol. 5, ssue 5, pp , 014. [] R. O. Preda, Sem-fragle atermarkng for mage authentcaton th senstve tamper localzaton n the avelet doman, Measurement, vol. 46, ssue 1, pp , 013.

14 1 Radu Ovdu Preda, Ioana Marcu, Amela Cobanu [3] S. Bravo-Soloro, L. Gan, A. K. Nand, and M. F. Aburdene, Secure prvate fragle atermarkng scheme th mproved tamperng localsaton accuracy, Informaton Securty, IET, vol. 4, pp , 010. [4] H. Kuo-Mng, C. Tng-Wen, S. Wen-Ka, and K. Cha-Nan, Automatc mage authentcaton and recovery usng multple atermarks, 01 8th Internatonal Conference on Informaton Scence and Dgtal Content Technology (ICIDT), pp , 01. [5] K. We-Chn, C. Te-Chh, W. Hsn-Lung, and C. Jen-Chun, A Fragle Watermarkng Scheme for Image Authentcaton th Tamper Detecton and Localzaton, Fourth Internatonal Conference on Genetc and Evolutonary Computng (ICGEC), pp , 010. [6] A. T. S. Ho, Z. Xunzhan, S. Jun, and P. Marzlano, Fragle Watermarkng Based on Encodng of the Zeroes of the Z-Transform, IEEE Transactons on Informaton Forenscs and Securty, vol. 3, pp , 008. [7] C. Qn, C.-C. Chang, and P.-Y. Chen, Self-embeddng fragle atermarkng th restoraton capablty based on adaptve bt allocaton mechansm, Sgnal Processng, vol. 9, pp , 01. [8] H. He, J. Zhang, and F. Chen, Adjacent-block based statstcal detecton method for selfembeddng atermarkng technques, Sgnal Processng, vol. 89, ssue 8, pp , 009. [9] A. M. Hassan, A. Al-Hamad, B. Mchaels, Y. M. Y. Hasan, and M. A. A. Wahab, Secure Self-Recovery Image Authentcaton Usng Randomly-Szed Blocks, 010 0th Internatonal Conference on Pattern Recognton (ICPR), pp , 010. [10] S. D. Ln, J. H. Ln, and C. Y. Chen, A ROI-based sem-fragle atermarkng for mage tamper detecton and recovery, Internatonal Journal of Innovatve Computng, Informaton and Control, vol. 7, pp , 011. [11] P. Korus, J. Balas, and A. Dzech, Toards Practcal Self-Embeddng for JPEG- Compressed Dgtal Images, Multmeda, IEEE Transactons on, vol. 17, pp , 015. [1] S. Agreste and L. Pucco, Wavelet-based atermarkng algorthms: theory, applcatons and crtcal aspects, Internatonal Journal of Computer Mathematcs, vol. 88, pp , 011. [13] R. Chamla, A. Khan, and I. Usman, Dual Watermarkng Method for Secure Image Authentcaton and Recovery, IEEE 13th Internatonal Multtopc Conference, INMIC 009, pp. 1-4, Dec [14] J. A. Mendoza-Norega, B. M. Kurkosk, M. Nakano-Mytake and H. Perez-Mean, Image Authentcaton and Recovery Usng BCH Error-Correctng Codes, Internatonal Journal of Computers, vol. 5, no. 1, pp. 6-33, 011. [15] X. Wang, D. Zhang, and X. Guo, Authentcaton and recovery of mages usng standard devaton, Journal of Electronc Imagng, vol. (3), 03301, 013. [16] R. Ullah, A. Khan, and A. S. Malk, Dual-purpose sem-fragle atermark: Authentcaton and recovery of dgtal mages, Computers and Electrcal Engneerng, vol. 39, pp , 013. [17] S. Som, S. Palt, K. Dey, D. Sarkar, J. Sarkar, and K. Sarkar, A DWT-based Dgtal Watermarkng Scheme for Image Tamper Detecton, Localzaton, and Restoraton, n Appled Computaton and Securty Systems. vol. 305, pp , 015.