Uploaded on

IJERA (International journal of Engineering Research and Applications) is International online, ... peer reviewed journal. For more detail or submit your article, please visit

IJERA (International journal of Engineering Research and Applications) is International online, ... peer reviewed journal. For more detail or submit your article, please visit

  • Full Name Full Name Comment goes here.
    Are you sure you want to
    Your message goes here
    Be the first to comment
    Be the first to like this
No Downloads


Total Views
On Slideshare
From Embeds
Number of Embeds



Embeds 0

No embeds

Report content

Flagged as inappropriate Flag as inappropriate
Flag as inappropriate

Select your reason for flagging this presentation as inappropriate.

    No notes for slide


  • 1. Megha Kansal, Sukhjeet k. Ranade, Amandeep Kaur / International Journal of Engineering Research and Applications (IJERA) ISSN: 2248-9622 Vol. 2, Issue4, July-August 2012, pp.1759-1763 Fragile Watermarking For Image Authentication Using A Hierarchical Mechanism Megha Kansal*, Sukhjeet K. Ranade**, Amandeep Kaur*** *(Department of Computer Science, Punjabi University, Patiala) ** (Department of Computer Science, Punjabi University, Patiala) *** (Department of Computer Science, Punjabi University, Patiala) ABSTRACT With ubiquitous computing, safeguarding II. BASICS ON DIGITAL WATERMARKINGthe creative content and intellectual property in a A digital image watermark is adigital form has become essential necessity. distinguishing piece of information that may beWatermarking is one such technique for protecting visible or invisible, is stuck to the data intended to bethe digital data like image, video and audio signal. protected. Digital watermarking is capable inIn this paper, we process a novel fragile copyright protection, data authentication, integritywatermarking using the Hierarchical Mechanism checks in multimedia contents [1]. A watermarkingwhich is a combination of Block-wise and Pixel- system is usually divided into three distinct steps,wise Approach. The embedded watermark data are embedding, attack and detection. In embedding, anderived both from pixels and blocks. On the algorithm accepts the host and the data to bereceiver side, one can first identify the blocks embedded and produces a watermarked signal. Thecontaining the tampered content, and then use the watermarked signal is then transmitted or stored,watermark hidden in the rest blocks to exactly usually transmitted to another person. If this personlocate the tampered pixels. By combining the makes a modification, this is called an attack. Thereadvantages of both block-wise and pixel-wise are many possible attacks. Detection is an algorithmtechniques, this scheme is capable of finding the which is applied to the attacked signal to attempt todetailed tampered positions even if the modified extract the watermark from it. If the signal was notarea is more extensive. Moreover, after localizing modified during transmission, then the watermark isthe tampered-pixel, the original watermarked still present and it can be extracted. If the signal isversion can be perfectly restored using exhaustive copied, then the information is also carried in theattempts. copy. The embedding takes place by manipulating the content of the digital data, which means theKeywords - Multimedia Security, Content information is not embedded in the frame around theAuthentication, Fragile Watermarking, data, it is carried with the signal itself. A secret key isTampered-pixel localization, Image restoration used during the embedding and the extraction process in order to prevent illegal access to the watermark asI. INTRODUCTION shown in Fig.1. The world has been witnessing a rapidgrowth of internet technologies, multimediadistribution and e-commerce for better quality, useand services. Because of these advancements, a largeamount of digital data is easily accessible to someindividual or group without the permission of theowner. This digital data can be easily manipulated,tampered and distributed with the help of powerfulimage processing tools. The digital data image, audio,video canbe altered, has needed a requirement fortechniques that decide integrity of the information.Cryptography was suggested as an effective tool toprevent illegal distribution. But in order to avoid thedigital access, a special hardware should be mergedwith cryptography tool, which make it costly.Authentication and recovery of tampered localization Fig. 1 Block Diagram of Watermarking Systemof a digital data are two critical requirements. So,watermarking technique is widely used to protect thedigital information. 1759 | P a g e
  • 2. Megha Kansal, Sukhjeet k. Ranade, Amandeep Kaur / International Journal of Engineering Research and Applications (IJERA) ISSN: 2248-9622 Vol. 2, Issue4, July-August 2012, pp.1759-1763III. CLASSIFICATION OF WATERMARKING IV. BACKGROUND OF FRAGILETECHNIQUES WATERMARKING TECHNIQUES Watermarking algorithms can be classified on A fragile marking system detects anyseveral criteria as given below: tampering in the marked image. Fragile Watermark is easily distorted when host image is modified by small  According to Visibility transformations. Block Diagram of Fragile i) Visible Watermark Watermarking is shown in Fig.2. ii) No Visible Watermark  According to ability of Watermark to resist attack i) Fragile Watermarking ii) Semi-fragile Watermarking iii) Robust Watermarking  According to domain of Watermark insertion and extraction i) Spatial domain Watermarking Technique ii) Transform domain Watermarking Technique  According to watermark detection and extraction i) Blind Watermarking ii) Non-blind Watermarking The non-blind watermarking requires thatoriginal image to exist for detection and extraction Fig. 2 Watermark embedding and Watermarkwhereas blind techniques do not require original detection process of Fragile Watermarking Systemimage. A visible watermark is easily detected by the Fragile watermarking systems are categorized intoobservation. A invisible watermark is not seen to the two categories according to the working and detected by the algorithms. İnvisible  Spatial Domain: The technique that workswatermark technique is broadly categorized into three directly on the pixel values of the hostclasses: Fragile, Semi-fragile and Robust image.watermarking.  Transform Domain: In this technique, Robust watermarking is used for copyright watermark is inserted into transformedprotection. The invisible robust watermark is coefficients of image.embedded in such a way that alterations made to thepixel value are perceptually not noticed and it can be First, fragile watermarking that worksrecovered only with appropriate decoding mechanism. directly in the spatial domain and second, that worksSemi fragile watermarking has features of both robust in a transform domain. Spatial domain techniquesand fragile watermarking. It is used for data embed the watermark in the least significant bit planeauthentication. It is sensitive to signal modification. for perceptual transparency. Transform domain alters the frequency coefficients of the data elements to hide Fragile watermarking is used for tamper the watermark data. Transform domain has proved todetection. The fragile watermark is a mark that is be more robust than spatial domain technique [2].readily altered or destroyed when a host image is Possible image transformations include the Discretemodified through a linear or non linear Fourier transformation DFT, Discrete cosinetransformation. The digital watermark is fragile to any transformation DCT, Discrete wavelet transformationkind of distortion; the watermark image may go DWT by modifying the coefficients of global or blockthrough. E.g.: the image after lossy compression transform. Most fragile watermarking systems embedprocessing could be found to be authentic by “robust” the mark in spatial domain described in Lin and Delpimage authentication but it would fail “fragile” image [3].authentication. For “fragile” image authentication,one bit error in a message leads to a totally different One of the first fragile watermarks forauthenticator. authentication was proposed by Walton [4]. In his scheme, key-dependent checksums of the seven most Fragile image authentication is highly significant bits of grayscales along pseudo-randomsensitive and dependent on the exact value of the walks hides in the least significant bits of pixelsimage pixels. forming the walk. First, in this scheme, attacker can modify image content while keep their LSB unchanged. Second, this cannot determine the exact regions of modification in verification process. Yeung 1760 | P a g e
  • 3. Megha Kansal, Sukhjeet k. Ranade, Amandeep Kaur / International Journal of Engineering Research and Applications (IJERA) ISSN: 2248-9622 Vol. 2, Issue4, July-August 2012, pp.1759-1763and Mintzer [5] proposed a scheme that authenticates locate the tampered pixels.individual pixels. This uses the secret key to generate The implementation process of fragile watermarkinga unique mapping that randomly assigns a binary can be divided into following stages asvalue to grey levels of the image. This scheme has 1. Watermark embedding in the imagevery good localization property but it is not easy to 2. Tampered- pixel localizationdevelop a feasible key management infrastructure 3. Restoration of original imagewithout introducing security gaps. 1. WATERMARK EMBED PROCEDURE Wong [6] described a scheme in which In the watermark embedding procedure, the 5image is divided into blocks and each block contains most significant-bit (MSB) planes in the host imagethe hash calculated from the MSBs in the LSBs of the are kept unchanged, and the 3 least-significant-bitpixels forming that block. The localization properties (LSB) planes are replaced with watermark data. Here,of this scheme are limited. Different attacks have been the watermark data are determined by the MSBs andproposed to either break or increase the security of made up of two parts, which are respectively used toWong‟s algorithm. A counterfeiting attack was identify tampered blocks and to locate tamperedproposed by Holliman and Memon [7]. This attack pixels.belongs to the class of vector quantization The detailed steps are as follows:counterfeiting and defeats any fragile technique Denote the numbers of rows and columns intargeting localization accuracy by watermarking small an original image as N1 and N2, the total number ofimage blocks independently. By making each pixels as N (N = N1 X N2), and the gray pixel-valueswatermark block dependent upon other blocks in Pn Є [0, 255], n = 1, 2… N. Each P n can bewatermarked image, the problem of watermark represented with 8 bits, Bn,7, Bn,6,….., Bn,0, wherecounterfeiting becomes infeasible. Another attackderived by Fridrich [8] that can be mounted against Bn,u = mod 2, u = 0, 1, . . . , 7the Yeung-Mintzer scheme is that the lookup table (1.1)and the binary logo can be inferred when the samelookup table and logo are reused for multiple images. For each pixel, generate M authentication bits Chang [9] proposed a block-based according to its 5most significant bits.watermarking scheme which divides the protectedimage into 3*3 image blocks and makes use of thecryptographic hash function for feature extraction ofthe blocks. The feature of image block is extractedfrom the eight neighboring pixels of the central pixelof the image block. The extracted feature is hidden (1.2)into the central pixel of that image block. This schemestill allows pixels to be tampered because the feature Where An are pseudo-random binaryof each image block does not depend on the central matrices derived from a secret key, and their size is Mpixel of the image block. The block features X 5. To ensure security, the matrices An should becomputed for the original central pixel and the mutually different. The arithmetic in (2) is modulo-2,tampered central pixel are the same. meaning that, if there is any change in the 5 MSBs of a pixel, the authentication bits will be flipped with a Zhang [10] used a scheme in which both probability1/2.pixel-derived and block-derived watermark data arecarried by LSBs of all bits. During the extraction of 1. According to a secret key, pseudo-randomly dividewatermark data, first blocks containing the tampered the M X N authentication bits into a series of subsets,data is identified then the watermark data hidden in each of which contains K bits. Then, calculatethe rest blocks are checked to identify the tampered modulus-2 sums of the K authentication bits in eachpixel. Disadvantage of this scheme is that if the subset, and call the (M x N/K) results the sum-bits.percentage of ratio between the numbers of tampered Here, we let M be a multiple of 5and K = 2M/5 so thatblocks and that of all blocks is more than 5% then this the number of sum-bits is 5N/2.will show that one half of the image is tampered butcannot locate the tampered pixels. 2. Assuming that both N1 and N2 are multiples of 8, we divide the original image into N/64 non-V. PROPOSED METHOD overlapped blocks sized 8x8. In each block, we This paper proposes the scheme in which the pseudo-randomly select 160 positions from the 3embedded watermark data are derived from pixels and LSB-layers according to the secret key. Also, theblocks. On the receiver side, one can first identify the LSB-selection in different blocks should be mutuallyblocks containing the tampered content, and then use different. Then, a total number of selected LSB isthe watermark hidden in the rest blocks to exactly 1761 | P a g e
  • 4. Megha Kansal, Sukhjeet k. Ranade, Amandeep Kaur / International Journal of Engineering Research and Applications (IJERA) ISSN: 2248-9622 Vol. 2, Issue4, July-August 2012, pp.1759-1763 5N/2, and replace the original bits at the selected The probability of a subset being usable for a certain positions with the sum-bits. pixel is 3. For each block, we collect the 320 original bits in the 5 MSB-layers and the 160 sum-bits used to (1.3) replace the selected LSBs. Then, feed the 480 bits into Then, for a given pixel in tampered blocks, a hash function to compute 32 hash-bits. Here, the the number of the usable subsets, nu, obeys the hash function must have the property that any change following distribution: on an input would result in a completely different output. Put the hash-bits into the 32 remaining positions in the 3 LSB-layers, and combine the (1.4) original MSBs and the substituted LSBs to produce a watermarked image. For each usable subset, we check whether or Authentication not the extracted sum-bit is consistent with the Watermarked MSB of bits modulus-2 sum of the pixel‟s authentication bit andhost image image Sum-bits other (K-1) authentication bits. If, and only if, the Authentication Sum-bit consistency is satisfied in all usable subsets, the pixel Data is judged as „„not tampered‟‟, indicating that there is bit generation generation combinatio no alteration in its 5 MSBs. Otherwise, it Is judged as n Hash a „„tampered‟‟ pixel. This way, a pixel without any MSB bits alteration in its 5 MSBs must be judged as „„not Block division Hash tampered‟‟, and probability with which a pixel function containing modified MSBs is falsely judged as „„not MSB tampered‟‟ is (1.5) Fig. 3 Watermark Embedding Procedure 3. RESTORATION OF ORIGINAL IMAGE 2. TAMPERED-PIXEL LOCALIZATION After finding a „„tampered‟‟ pixel, we can Assume that an attacker may alter the gray further recover its original MSBs. The number of values of some pixels without changing the image possible patterns of 5 MSBs is 32. We attempt to use size. After receiving the image, we want to locate the 31 other patterns different from the received pattern of tampered pixels and restore the original content. Here, the pixel to check consistency between the extracted „„tampered pixels‟‟ are those with changes in their 5 sum-bit and the modulus-2 sum of the pattern‟s MSBs. The tampered-pixel localization procedure is authentication bit and other (K-1) authentication bits. made up of two stages. When consistency is arrived in all usable subsets, the attempted pattern is regarded as the original MSBs. If 1. The first is to identify the tampered blocks. After more than one pattern satisfies the consistency dividing the received image into non-overlapped 8x8 condition in all usable subsets, restoration of original blocks, we select 160 positions from the 3 LSB-layers pattern will be failed since we do not know which one in each block according to the same secret key. For is the true original pattern. The true original pattern each block, if the hash result of the 320 bits in the 5 must satisfy the consistency condition in all usable MSB-layers and the 160 bits at the selected positions subsets, and the probability of the other pattern in 3 LSB-layers is identical to the 32 bits at the other satisfies the consistency condition in all usable LSB positions, the block is judged as “not tampered‟‟. subsets is 2-nU. So, probability of failure is 1-(1-2- nU 30 2. In the second stage, we locate the tampered pixels ) . Considering the distribution of nu, probability in the tampered blocks. Considering a pixel in for the original MSBs of a pixel with „„tampered‟‟ tampered blocks, its M authentication bits are judgment being unable to find is distributed into M subsets, each of which contains K elements. For each subset, if all the other (K-1) (1.6) authentication bits in it are derived from pixels in „„not tampered‟‟ blocks and its sum-bit is also hidden Denoting the number of tampered pixels in a „„not tampered‟‟ block, we say the subset is labeled „„tampered‟‟ but being unable to be restored usable for the pixel. So, a receiver can derive the (K- as NC, its average is 1) authentication bits in usable subset from their corresponding pixels, and extract the sum-bit from its embedding position. Denote a ratio between the (1.7) numbers of tampered blocks and that of all blocks as α. 1762 | P a g e
  • 5. Megha Kansal, Sukhjeet k. Ranade, Amandeep Kaur / International Journal of Engineering Research and Applications (IJERA) ISSN: 2248-9622 Vol. 2, Issue4, July-August 2012, pp.1759-1763If NC = 0, the receiver can obtain the original MSBs of block is identified then the watermarks hidden in theall pixels, leading to restoration of the original rest blocks are used to exactly locate the tamperedwatermarked image without any error. pixel and then restore the original watermarked version. By using Gaussian filter we get the betterVI. PERFORMANCE EVALUATION OF DIGITAL image with fragile watermark. So, improves theIMAGE WATERMARKING ALGORITHM PSNR value. This scheme is also capable of recovering the original content and regenerating the Performance evaluation of the algorithm is watermarked version on the receiver side. Thisdone with the perceptual transparency. Perceptual scheme have the property of both the methods block-transparency means the quality of the image should wise and pixel-wise fragile watermarking.not be destroyed by the presence of watermark. The In future work, we eliminate thequality of the watermark is measured by PSNR (Peak disadvantage of [10]. So, in comparison with schemeSignal to Noise Ratio). Bigger the PSNR better the described in [10], the proposed scheme will exactlyquality of watermarked image. locate the tampered pixel if the percentage of ratioPSNR is measued with the formula: between the numbers of tampered blocks and that of all blocks is more than 5%.(1.8) REFERENCES Where, EQM is the quadratic mean error [1] N. Memon and P. W. Wong, “Protectingbetween two images, I be an input image and be an digital media content: Watermarks forapproximation that result from digital image copyrighting and authentication,”watermarking. İmages are of size N*M. Communications of ACM, July 1998. [2] I. Cox, J. Kilian and T. Shamoon, “SecureEQM = Spread Spectrum Watermarking for(1.9) Multimedia”, IEEE Transaction on Imagewhere, are the intensities of gray Processing, vol. 6, Dec. 1997, 1673-1687. [3] Lin and E. Delp, “A Review of Fragilelevel images I and at position (i,j). Image Watermarks,” in Proc. of the Multimedia and Security Workshop, 1999, İn this paper to improve the picture quality of 25-29.restored image we use Gaussian filter. This filter does [4] S. Walton, “Information authentication for asharpening and smoothing of image gray value. By slippery new age”, Dr. Dobbs J., vol. 20,using Gaussian filter we get better image with fragile 1995, 18–26.watermark. As we know bigger the PSNR, better the [5] M. M. Yeung and F. Mintzer, “An invisiblequality watermarked image. watermarking technique for image verification”, in Proc. ICIP, Santa Barbara, Where, M is the authentication bits for each CA, 1997.pixel according to its 5most significant bits. [6] P. W. Wong, “A public key watermark forThe results are as follows: image verification and authentication”, in Proc. ICIP, Chicago, IL, Oct. 1998. Value of M PSNR of PSNR of [7] M. Holliman and N. Memon, “Counterfeiting watermarked Gaussian attacks on oblivious block-wise independent image(dB) filtered image invisible watermarking schemes”, IEEE 40 37.87 39.89 Trans. Image Processing, vol. 9, 2000, 432– 441. 50 38.03 39.91 [8] M. Fridrich and A. Baldoza, “New Fragile Authentication Watermark For Images,” in 60 38.35 39.96 Proc. of the IEEE International Conference on Image Processing, Vancouver, Canada,TABLE showing the comparsions of PSNR values of 2000, 446–449.Watermarked image with Gaussian filtered image. [9] C. C. Chang, Y. S. Hu, and T. C. Lu, “A watermarking-based image ownership andVII. CONCLUSION AND FUTURE WORK tampering authentication scheme”, Pattern Recognition Letters, vol. 27, April 2006,The proposed fragile watermarking is based on 439-446.hierarchical mechanism in which watermark data [10] X. Zhang and S. Wang, “Fragilederived from the MSBs and directly replace all the Watermarking scheme using a hierarchicalLSBs of a host image. Watermarking embedding mechanism”, Signal Processing, vol. 89,procedure is based on spatial domain. So, it‟s easy to 2009, 675-679.implement. On the receiver side, first the tampered 1763 | P a g e