Welcome to Scribd, the world's digital library. Read, publish, and share books and documents. See more
Download
Standard view
Full view
of .
Save to My Library
Look up keyword
Like this
3Activity
0 of .
Results for:
No results containing your search query
P. 1
An Efficient Vector Quantization Method for Image Compression with Codebook generation using Modified K-Means

An Efficient Vector Quantization Method for Image Compression with Codebook generation using Modified K-Means

Ratings: (0)|Views: 284 |Likes:
Published by ijcsis
With the growth of internet and multimedia, compression techniques have become the thrust area in the fields of computers. Image compression is a technique of efficiently coding digital image to reduce the number of bits required in representing image. Many image compression techniques presently exist for the compression of different types of images. In this paper Vector Quantization based compression scheme is introduced. In this scheme a low bit rate still image compression is performed by compressing the indices of Vector Quantization and residual codebook is generated. The indices of VQ are compressed by exploiting correlation among image blocks, which reduces the bit per index. A residual codebook similar to VQ codebook is generated that represents the distortion produced in VQ. Using this residual codebook the distortion in the reconstructed image is removed, thereby increasing the image quality. The proposed technique combines these two methods and by replacing the Modified k-means algorithm for LBG in the codebook generation. Experimental results on standard image Lena show that the proposed scheme can give a reconstructed image with a higher PSNR value than all the existing image compression techniques.
With the growth of internet and multimedia, compression techniques have become the thrust area in the fields of computers. Image compression is a technique of efficiently coding digital image to reduce the number of bits required in representing image. Many image compression techniques presently exist for the compression of different types of images. In this paper Vector Quantization based compression scheme is introduced. In this scheme a low bit rate still image compression is performed by compressing the indices of Vector Quantization and residual codebook is generated. The indices of VQ are compressed by exploiting correlation among image blocks, which reduces the bit per index. A residual codebook similar to VQ codebook is generated that represents the distortion produced in VQ. Using this residual codebook the distortion in the reconstructed image is removed, thereby increasing the image quality. The proposed technique combines these two methods and by replacing the Modified k-means algorithm for LBG in the codebook generation. Experimental results on standard image Lena show that the proposed scheme can give a reconstructed image with a higher PSNR value than all the existing image compression techniques.

More info:

Published by: ijcsis on Dec 04, 2010
Copyright:Attribution Non-commercial

Availability:

Read on Scribd mobile: iPhone, iPad and Android.
download as PDF, TXT or read online from Scribd
See more
See less

12/04/2013

pdf

text

original

 
 Abstract—
With the growth of internet and multimedia,compression techniques have become the thrust area in the fields of computers. Image compression is a technique of efficiently codingdigital image to reduce the number of bits required in representingimage. Many image compression techniques presently exist for thecompression of different types of images. In this paper VectorQuantization based compression scheme is introduced. In thisscheme a low bit rate still image compression is performed bycompressing the indices of Vector Quantization and residualcodebook is generated. The indices of VQ are compressed byexploiting correlation among image blocks, which reduces the bit perindex. A residual codebook similar to VQ codebook is generated thatrepresents the distortion produced in VQ. Using this residualcodebook the distortion in the reconstructed image is removed,thereby increasing the image quality. The proposed techniquecombines these two methods and by replacing the Modified k-meansalgorithm for LBG in the codebook generation. Experimental resultson standard image Lena show that the proposed scheme can give areconstructed image with a higher PSNR value than all the existingimage compression techniques.
 Keywords—
Image compression, Vector Quantization, ResidualCodebook, Modified K-Means
 I.
 
I
NTRODUCTION
 MAGE compression is a method of efficiently codingdigital image, to reduce the number of bits required inrepresenting image. Its purpose is to decrease the storagespace and transmission cost while maintaining good quality.VECTOR Quantization [1] has been found to be an efficienttechnique for image compression in the past decade. VQcompression system mainly contains two components: VQencoder and decoder as shown in Fig.1.In VQ technique [2] [3], the input image is partitioned intoa set of non-overlapping image blocksof size 4x4 pixels each and a clustering algorithm, for exampleLinde–Buzo–Gray (LGB) algorithm [5] and Modified K-Means [2]. The Modified K-Means algorithm is used in theproposed technique, to generate a codebook for the given set of image blocks. Thecodebook C comprises a set of representative image blockscalled codewords. The VQ encoder discovers a closest matchcodeword in the codebook for each of the image block and theindex of the codeword is transmitted to VQ decoder. Thedecoder phase has the following functionalities. VQ decoderreplaces the index values with the respective codewords fromthe codebook and produces the quantized image, called asreconstructed image. In order to attain low bit rate, many VQschemes, have been used in the past literature such as side-match VQ (SMVQ) [6], classified SMVQ (CSMVQ) [7] andGradient based SMVQ (GSMVQ) [8].(a)(b)Fig. 1 (a) VQ Encoder (b) VQ DecoderEven though, SMVQ, CSMVQ, GSMVQ and JSPVQ arethe low bit rate schemes, they require high encoding time thanthat of VQ method. In this paper, an efficient low bit rateimage compression scheme is proposed based on VQ thatmakes use of compression of indices of VQ and residualcodebook with modified k-means clustering instead of LGB.This scheme attains low bit rate and better image quality thanprevious VQ methods.The rest of the paper is presented as follows: in section II,the literature survey is presented. In section III the proposedcompression scheme is described. Performance of theproposed system is evaluated in section IV and section Vconcludes the paper.
An Efficient Vector Quantization Method forImage Compression with Codebook generation usingModified K-Means
S. Sathappan, Associate Professor of Computer Science, Erode Arts and Science College, Erode-638 009.Tamil Nadu. India. E-Mail : devisathappan@yahoo.co.in
I
Codebook C={c
1
,c
2
..c
n
}VQEncoderLBG / ModifiedK-MeansIndicesImage BlocksCodebook C={c
0
,c
2
..c
N-1
}VQDecoderLBG / ModifiedK-MeansReconstructed ImageIndices
(IJCSIS) International Journal of Computer Science and Information Security,Vol. 8, No. 8, November 2010206http://sites.google.com/site/ijcsis/ISSN 1947-5500
 
II.
 
L
ITERATURE
S
URVEY
 Somasundaram et al, [1] presented a novel vectorquantization approach and also explained various VQschemes, such as side-match VQ (SMVQ), classified SMVQ(CSMVQ) and Gradient based SMVQ (GSMVQ). In whichSMVQ uses the high correlation existing between neighboringblocks to achieve low bit rate and master codebook C is usedto encode image blocks in the first column and first row inadvance. The other image blocks are encoded, utilizing thecorrelation with the neighboring encoded image blocks. Let xbe the input image block for the compression system, and uand l be the upper and left neighboring codewordsrespectively. Let the size of the given image block size be k =m× n. The side-match distortion of a codeword Y can bedefined as:(1)According to their side-match distortions of all codewordsSMVQ sorts the codewords and then selects N
S
codewordswith smallest side-match distortions from the master book Cof size N to form the state codebook SC, where N
S
< N. Abest-match codeword Y
i
is selected to encode an image block x from N
S
codewords and the corresponding index is coded inlog
2
N
S
bits. Thus, the SMVQ reduces the bit rate of VQ. Sincemean square error caused by state codebook is higher than thatof master codebook, SMVQ degrades the image quality andalso it requires long encoding time. Classified side-matchvector quantization [7] (CSMVQ) is an efficient low bit rateimage compression technique which produces relatively highquality image. It is a variable rate SMVQ and makes use of variable sized state codebooks to encode the current imageblock. The size of the state codebook is decided based on thevariances of left codewords and upper codewords that predictthe block activity of the input blocks. Also, CSMVQ uses twomaster codebooks, one for low detail blocks and another forhigh detail blocks. Another variant, gradient-based SMVQ [8](GSMVQ) has been proposed, in which gradient values areused instead of variance values to predict the input vector.Another low bit rate VQ, called Jigsaw-puzzle vectorquantization (JPVQ) [9] was proposed,in which an input block can be coded by the super codebook, the dynamic codebook orthe jigsaw-puzzle block. The jigsawpuzzle block isconstructed dynamically using four-step side match predictiontechnique.Interpolative vector quantization, first proposed explicitlyby Gersho in [12], introduces dimension reduction totraditional VQ. The codebook in the encoder is learned ondownsampled vectors and the codebook in the decoder onhigh-dimension vectors. Except for the difference ondimension, the two codebooks have the same number of representative vectors and structure. VQ encoder maps downthe sampled inputs to a set of scalar indices and VQ decoderreproduces high-dimension inputs by received indices. Davidet al. applied IVQ to image restoration [13], where the encoderdoes not need a codebook except for some parameters. Thecodebook at decoder is learned on image pairs consisting of anoriginal image and its diffraction-limited counterpart. Severalfollow-up work is reported in [14][15].The goal of quantization is to encode the data from asource, with some loss, so that the best reproduction isobtained. Vector quantization (VQ) achieves morecompression then scalar quantization [14], making it useful forband-limited channels. The algorithm for the design of optimalVQ is commonly referred to as the Linde-Buzo-Gray (LBG)algorithm, and it is based on minimization of the squared-errordistortion measure. The LBG algorithm starts with an initialcodebook and iteratively partitions the training sequence intothe Voronoi regions to obtain a new codebook that produces alower distortion. Once the final codebook is got, it can be usedon new data outside the training sequence with the optimumnearest neighbor rule. If the training sequence is sufficientlylong, it yields good performance for future data produced bythe source.In the paper by M.Antonini, et al. [18], images have beencoded using two-level wavelet decomposition with VQ of theresulting coefficients. A multiresolution codebook has beendesigned with noise-shaping bit allocation among the varioussubbands. The test image has been coded at a rate of 0.78bpp,achieving a PSNR of 32.1dB. In the paper by Gersho andRamamurthy [19], images have been compressed usingunstructured VQ, achieving a bitrate of 0.5 – 1.5bpp. Ho andGersho [19] have used multistage VQ for progressive imagecoding, with a PSNR of 30.93dB at 0.36bpp using 4 stages.R.L.Claypoole et al. [21] have coded images using nonlinearwavelet transform via lifting and obtained 30.5dB at 0.6bpp.An adaptive lifting scheme with perfect reconstruction hasbeen used in [21].III.
 
M
ETHODOLOGY
 The compression scheme consists of two components,compression of indices and generation of residual codebook.These two are explained in this section.
3.1.
 
Compression of Indices
The Index compression step has the following procedure.When the image blocks are vector quantized, there likely toexist high correlation among the neighboring blocks and henceamong the corresponding codeword indices. Therefore, if indices are coded by comparing with the previous indices,further reduction in the bit rate can be achieved. In SearchOrder Coding (SOC) [11], a simple searching scheme isfollowed to find a match for the current index from theprevious indices. The search order SO is defined as the orderin which the current index is compared with the previousindices.
(IJCSIS) International Journal of Computer Science and Information Security,Vol. 8, No. 8, November 2010207http://sites.google.com/site/ijcsis/ISSN 1947-5500
 
 Fig. 2 Searching OrderThe SO used in [11] is given in Fig.2. The label “1”indicates the highest searching priority, “2” denotes the secondhighest priority and so on. In order to limit comparisons of thecurrent index with previous indices, the searching range (SR)is fixed. The SR is defined as the number of previous indicesto be compared with current index. In this paper Search Order[11], SR is taken as 10, which gives the lower bit rate thanother SR.In this method, the index of the codeword of a block isencoded exploiting the degree of the similarity of the block with previously encoded upper or left blocks. When the degreeof similarity of the current block with one of the twopreviously encoded blocks is high, the index of the codewordof the block is encoded using the index of the neighboringcodeword. I.e. the codeword index of the current block andthat of the neighboring blocks are same. If the degree of similarity of the block with the neighboring blocks is not high,it is assumed that the closest match codeword of the currentblock may be nearer to the codewords of the neighboringblocks. For example, if one of the two neighboring blockscodeword’s index is ‘N’, the closest match codeword of theblock to be encoded may lie between (N-J) th codeword and(N+J) th codeword in the codebook, where J is any arbitrarynumber. So the index can be coded in log
2
( 2*J ) bits. Thisidea is based on the property existing in the codebook designusing LBG algorithm with splitting technique. In the splittingtechnique, bigger size codebook is generated by splitting eachcodeword of the smaller codebook into two. The size of thecodebook is always in powers of two (2
M
 
2
(M+1)
). Hence,relatively similar two image blocks may have same closestmatch codeword in Jth position at codebook of size 2
M
and atcodebook of size 2
(M+1)
, one of the two image blocks may haveits closest match codeword at Jth place in the codebook andother block’s codeword may be in (J+1)th place. The othernon-similar blocks are encoded using their original indexvalue. In this scheme, examining the resemblance of a block with its left and upper blocks is not required to encode theindex of the block. The above description is the idea behindour VQ indices compression scheme. In order to implementthis idea, the index to be coded is compared with previousindices according to the SO given in Fig.2 and SR is fixed as 2in this scheme. Let 1, 2,..,12 be the SO and ind_val (1),ind_val (2),..ind_val(12) be the indices values of the SO =1,2,…12. The following steps is used to encode VQ index.1 Get the first index generated by the VQ encoder and transmitas such.2. Get the next index generated by VQ Encoder. Compare thisindex with the previous indices according SO3. if SO = 1, code it as “00” and go to the step 2elseif SO = 2, code it as “01” and go to the step 2else go to the next step.4 if index value
(ind_val (SO = 1) + J) andindex value
 
 ─ 
(ind_val (SO = 1)+J){ if ind_val (SO =1) = ind_val (SO=2)code it as “10” followed by log (2 * ) 2 J bitselsecode it as “100” followed by log (2 * ) 2 Jbits. }go to step 2.
else
if index value
(ind_val (SO = 2) + J) andindex value
 
 ─ 
(ind_val (SO = 2)+J)code it as “101” followed by log (2 * ) 2 J bitsand go to step 2.elsecode it as “11” followed by its original index andgoto step 2.Decoding of the compressed indices is done by reversing theabove coding steps.
3.2.
 
Construction of Residual Codebook (RC)
The residual codebook can be representedas. Residual codebook isconstructed using absolute error values caused by VQ method,in the residual codebook construction, the image blocks thatare less similar to their closest match codewords found in thecodebook are taken into account. Less similarity blocks willincrease distortion than high similarity blocks in thereconstructed image. Residual codeword (RY
i
) for a lesssimilarity image block is constructed by comparing it with itsclosest match codeword. The collection of residual codewordsRY
i
, RY
i+1
… is called residual codebook. Similarity of animage block x with its closet match codeword Y
i
is determinedbased on minimum distortion rule (
α
) between them. If themean square error (
α
) of an image block is greater than apredefined threshold value (
σ
), then the block is taken as lesssimilarity block.Let be a k-pixels image block andbe a k-pixels closest match codeword,then the
α
is defined as:(2)The steps used for constructing residual codebook aregiven below.8 6 9 127 3 2 4 105 1
 
Currentindex
(IJCSIS) International Journal of Computer Science and Information Security,Vol. 8, No. 8, November 2010208http://sites.google.com/site/ijcsis/ISSN 1947-5500

Activity (3)

You've already reviewed this. Edit your review.
1 thousand reads
1 hundred reads
Akshay Kale liked this

You're Reading a Free Preview

Download
/*********** DO NOT ALTER ANYTHING BELOW THIS LINE ! ************/ var s_code=s.t();if(s_code)document.write(s_code)//-->