(IJCSIS) International Journal of Computer Science and Information Security,Vol. 8, No. 2, 2010
Comparative Performance of InformationHiding in Vector Quantized Codebooks usingLBG, KPE, KMCG and KFCG
Dr. H. B. Kekre
Senior Professor,MPSTME,NMIMS University,Vile-parle(W),Mumbai-56, India.hbkekre@yahoo.com
Archana Athawale
Ph.D. Scholar, MPSTME,NMIMS University,Vileparle(W), Mumbai-56Assistant Professor, TSEC,Bandra(W), Mumbai-50,India.
Tanuja K. Sarode
Ph.D. Scholar, MPSTME,NMIMS University,Vileparle(W), Mumbai-56Assistant Professor, TSEC,Bandra(W), Mumbai-50,India.
Kalpana Sagvekar
Lecturer,Fr. Conceicao RodriguesCOE, Bandra(W),Mumbai-50, Indiakalpanasagvekar@gmail.com
Abstract -
In traditional VQ - data hiding schemes secret data ishidden inside index based cover image resulting in limited embeddingcapacity. To improve the embedding capacity as well as to haveminimum distortion to carrier media, we have proposed one novelmethod of hiding secret data into the codebook. In this paper we have
used four different algorithms Linde Buzo and Gray (LBG), Kekre’s
Proportiona
te Error (KPE), Kekre’s Median Codebook Generationalgorithm (KMCG) and Kekre’s Fast Codebook Generation Algorithm
(KFCG) to prepare codebooks. It is observed that KFCG givesminimum distortion.
Keywords
-
Reversible (lossless) data hiding, VQ, LBG, KPE, KMCG,KFCG
.
I.
INTRODUCTIONDue to the digitalization of all kinds of data and the amazingdevelopment of network communication, information securityover the Internet has become more and more important. TheInternet is basically a giant open channel with securityproblems like modifications and interceptions occurring at anytime in any place. Under such circumstances, quite somedifferent approaches have been proposed in an attempt to makeprivate communication secure. Researchers have developedschemes where the secret message is protected by gettingtransformed into the form of a stack of seemingly meaninglessdata, which only the authorized user can retransform back to itsoriginal form by way of some secret information. However, theappearance of a stack of seemingly meaningless data could bean irresistible attraction to an attacker with a desire to recoverthe original message. Another approach, called steganography,hides the secret message in some cover material with acommon appearance to avoid suspicion. The data-hidingefficacy can be judged according to two criteria: (1) visual
quality (2) payload capacity limit. The term “visual quality”
here refers to the quality of the stego-image. That is to say, onlya limited number of distortions within limited areas are allowedin the stego-image so that no obvious traces of modification
appear on the picture to catch malicious attackers‟ attention.
Thereupon, the security of the secret information is ensuredagainst detection. As for the payload capacity limit, itevaluates the power of a data-hiding scheme by checkinghow big the maximum amount of the secret information isthat can be hidden in the cover media. Generally speaking,the larger the payload size is, the worse the stego-imagevisual quality will be. That is to say, in the world of datahiding, how to strike this balance and settle on an idealrobustness-capacity tradeoff is maybe the core problem tosolve.The existing schemes of data hiding can roughly be classifiedinto the following three categories:Spatial domain data hiding [2],[3],[4]: Data hiding of thistype directly adjust image pixels in the spatial domain fordata embedding. This technique is simple to implement,offering a relatively high hiding capacity. The quality of thestego image can be easily controlled. Therefore, data hidingof this type has become a well known method for imagesteganography.Frequency domain data hiding [5],[6]: In this method imagesare first transformed into frequency domain, and then data isembedded by modifying the transformed coefficients.Compressed domain data hiding [7],[8]: Data hiding isobtained by modifying the coefficients of the compressedcode of a cover image. Since most images transmitted overInternet are in compressed format, embedding secret data intothe compressed domain would provoke little suspicion.Due to the restricted bandwidth of networks, we cannot keepup with the growing sizes of various multimedia files. Manypopular image compression algorithms have been proposedto respond this problem, such as VQ [15], side match VQ(SMVQ) [16], JPEG [17], JPEG2000 [18], and so on. One of the most commonly studied image compression techniques isVector Quantization (VQ) [19], which is an attractive choice
89http://sites.google.com/site/ijcsis/ISSN 1947-5500
(IJCSIS) International Journal of Computer Science and Information Security,Vol. 8, No. 2, 2010
because of its simplicity and cost-effective implementation.Indeed, a variety of VQ techniques have been successfullyapplied in real applications such as speech and image coding[20], [22], VQ has faster encode/decode time along withsimpler framework compared to JPEG/JPEG2000. VectorQuantization requires limited information during decoding andworks best in applications in which the decoder has onlylimited information [21].There are two approaches for hiding data into VQ compresseddomain; either hides the covert data into index based coverimage or in codebook. In this paper we have proposed a methodof hiding data into codebook which is not been explored. Insection II we present codebook design algorithms. Section IIIexplains proposed search algorithm followed by Section IV inwhich results and evaluation is given. Section V givesconclusion.II. VQ COMPRESSION TECHNIQUEVector Quantization (VQ) [9-14] is an efficient technique fordata compression [31-34] and is very popular in a variety of research fields such as data hiding techniques [7,8], imagesegmentation [23-26], speech data compression [27], contentbased image retrieval CBIR [28, 29] and face recognition [30].
A. Codebook Generation Algorithmsa. Linde-Buzo-Gray (LBG) Algorithm [9], [10]
In this algorithm centroid is calculated as the first codevectorfor the training set. In Fig. 1 two vectors v1 & v2 are generatedby using constant error addition to the codevector. Euclideandistances of all the training vectors are computed with vectorsv1 & v2 and two clusters are formed based on nearest of v1 orv2. This procedure is repeated for every cluster. The drawback of this algorithm is that the cluster elongation is
–
45
o
tohorizontal axis in two dimensional cases. Resulting ininefficient clustering.Fig.1 LBG for 2 dimensional case
b. Proportionate Error Algorithm (KPE) [11], [12]
Here proportionate error is added to the centroid to generatetwo vectors v1 & v2. Magnitude of elements of the centroiddecides the error ratio. Hereafter the procedure is same asthat of LBG. While adding proportionate error a safe guard isalso introduced so that neither v1 nor v2 go beyond thetraining vector space. This removes the disadvantage of theLBG. Both LBG and KPE requires 2M number of Euclideandistance computations and 2M number of comparisons whereM is the total number of training vectors in every iteration togenerate clusters.
c.
Kekre’s Median Codebook Generation Algorithm
(KMCG) [13]
In this algorithm image is divided in to blocks and blocks areconverted to the vectors of size k. The Fig. 2 belowrepresents matrix T of size M x k consisting of M number of image training vectors of dimension k. Each row of thematrix is the image training vector of dimension k.
kM,x....M,2xM,1x........k2,x....2,2x2,1xk1,x....1,2x1,1xT
Fig. 2. Training VectorsThe training vectors are sorted with respect to the firstmember of all the vectors i.e. with respect to the first columnof the matrix T and the entire matrix is considered as onesingle cluster. The median of the matrix T is chosen(codevector) and is put into the codebook, and the size of thecodebook is set to one. The matrix is then divided into twoequal parts and the each of the part is then again sorted withrespect to the second member of all the training vectors i.e.with respect to the second column of the matrix T and weobtain two clusters both consisting of equal number of training vectors. The median of both the parts is the pickedup and written to the codebook, now the size of the codebook is increased to two consisting of two codevectors and againeach part is further divided to half. Each of the above fourparts obtained are sorted with respect to the third column of the matrix T and four clusters are obtained and accordinglyfour codevectors are obtained. The above process is repeatedtill we obtain the codebook of desired size. Here quick sortalgorithm is used and from the results it is observed that thisalgorithm takes least time to generate codebook, sinceEuclidean distance computation is not required.
d.
Kekre’s Fast
Codebook Generation (KFCG) Algorithm
In [14], KFCG algorithm for image data compression isproposed. This algorithm reduces the time for codebook generation. It does not use Euclidian distance for codebook generation. In this algorithm image is divided in to blocksand blocks are converted to the vectors of size k. Initially wehave one cluster with the entire training vectors and thecodevector C1 which is centroid.
90http://sites.google.com/site/ijcsis/ISSN 1947-5500
(IJCSIS) International Journal of Computer Science and Information Security,Vol. 8, No. 2, 2010
In the first iteration of the algorithm, the clusters are formed bycomparing first element of training vector with first element of code vector C1. The vector X
i
is grouped into the cluster 1 if x
i1
< c
11
otherwise vector X
i
is grouped into cluster 2 as shownin Figure. 3a. where codevector dimension space is 2.In second iteration, the cluster 1 is split into two by comparingsecond element x
i2
of vector X
i
belonging to cluster 1 with thatof the second element of the codevector which is centroid of cluster 1. Cluster 2 is split into two by comparing the secondelement x
i2
of vector X
i
belonging to cluster 2 with that of thesecond element of the codevector which is centroid of cluster2, as shown in Figure. 3b.This procedure is repeated till the codebook size is reached tothe size specified by user. It is observed that this algorithmgives less error as compared to LBG and requires least time togenerate codebook as compared to other algorithms, as it doesnot require computation of Euclidian distance.3(a). First Iteration3(b) Second IterationFig. 3. KFCG algorithm for 2-D case
III.
PROPOSED APPROACH
In this approach, we are hiding the secret data into codebook generated using various codebook generation algorithm such asLBG[10][11], KPE[12][13], KMCG[14], KFCG[15]. There arevarious ways of hiding: 1bit, 2 bits, 3 bits, 4 bits & variable bitshiding.
A.
Embedding Procedure B.
Extraction & Recovery ProcedureC. Variable Bit Hiding Algorithm
For variable bit hiding Kekre‟s algorithm [2] is used.
1.
If the value of codebook vector element is in the range
240≤g
i
≤255 then we embed 4 bits of secret data into the4 LSB‟s codebook vector element. This can be done byobserving the 4 most significant bits (MSB‟s). If they areall 1‟s then the remaining 4 LSB‟s can be used for
embedding data.2.
If the value of codebook vector element is in the range
224≤gi≤239 then we embed 3 bits of secret data. . This
Divide the image into 2×2 block of pixelsGenerate initial cluster of training set using therows of 12 values per pixel windowApply codebook generation algorithmLBG/KPE/KFCG/KMCG on initial cluster toobtain codebook of size 2048 codevectorsEmbed every bit of each pixel in the LSB
‟s
of (i.e. 1, 2, 3, 4, variable bit method) each elementof codevector belonging to CBGenerate Index based cover imageModified CBModified CBExtract secretdata fromLSB of everyelement of CBReconstruct the originalimage by replacingeach index bycorrespondingcodevectorIndex based cover image
91http://sites.google.com/site/ijcsis/ISSN 1947-5500
Search History:
Result 00 of 00
00 results for result for
p.
Comparative Performance of Information Hiding in Vector Quantized Codebooks using LBG, KPE, KMCG and KFCG
In traditional VQ - data hiding schemes secret data is hidden inside index based cover image resulting in limited embedding capacity. To improve the embedding capacity as well as to have minimum distortion to carrier media,…
(More)
751
Reads
4
Readcasts
3
Embed Views
More From This User
Notes
Load more
