Improved Error Detection in the JPEG2000 Codec

Transcription

1 Improved Error Detecton n the JPEG2000 Codec Farzad Zargar, Member, IEEE, Al Sefdpour, M. Shahram Mon, Member, IEEE, and Mohammad Ghanbar, Fellow, IEEE Abstract the Segmentaton symbol s an error reslence feature n the JPEG2000 mage codng standard. When ths feature s used, lack of detecton of the segmentaton symbol at the end of a bt-plane by the decoder ndcates that an error has occurred wthn that bt-plane. However, a detected and correctly decoded segmentaton symbol does not guarantee that the bt-plane s error free. In ths paper, an error detecton method s proposed for detectng and removng the artfacts caused by the damaged bt-planes that were not detected by the segmentaton symbol. The proposed method provdes a model for the coded bt-planes n the encoder and uses ths model for error detecton at the decoder. Extensve smulatons verfy that the proposed method can effectvely remove artfacts and mprove the qualty of the corrupted mages on average by 1.2 db and n some pctures by up to 6 db n PSNR 1. Key words JPEG2000, Error Detecton, segmentaton symbol, decodng artfacts. I. INTRODUCTION Multmeda commutatons n wreless and broadcast networks are becomng ncreasngly mportant and are wdely used n a varety of applcatons. Dgtzed mages are among the most commonly used data n multmeda communcatons. The delvery path for dgtzed mages may nclude dfferent parts, wth varous bandwdth and Qualty of Servces (QoS) and may mpose dverse losses on the transmtted data. Hence, the man concerns n the wred or wreless multmeda applcatons are the protecton of transmtted data aganst varous channel errors and recovery of the erroneous data. Hence, the mage and vdeo codng standards provde error reslence (ER) tools to protect the coded data aganst channel error n error prone channels. Although the provded ER tools mprove the decodng of erroneous data, they can not detect all the errors that may appear n the coded bt stream. The undetected bt errors on the one hand may result n artfacts n the decoded mage or vdeo and on the other hand they may degrade the performance of error concealment technques whch assume that errors are correctly located. Thus, many researchers have been workng on error detecton tools for detectng the errors whch cannot be detected by the 1 F. Zargar s wth the Iran Telecom Research Center (ITRC), Tehran 14395, Iran (e-mal: zargar@trc.ac.r). A. Sefdpour s wth Iran Islamc Azad Unversty Sc. & Res. Campus, Tehran, Iran (e-mal: sefdpour@pcd.r). M. S. Mon s wth the Iran Telecom Research Center (ITRC) Tehran 14395, Iran (e-mal: mon@trc.ac.r). M. Ghanbar s wth the Department of Computng and Electronc Systems, Unversty of Essex, Colchester, Unted Kngdom (e-mal: ghan@essex.ac.uk). ER tools n the standard codecs. The reported error detecton tools n the lterature can be classfed nto two man groups. The methods n the frst group employ the avalable nformaton of the coded mage or vdeo for error detecton and n the second group they rely on the addtonal nformaton ncluded n the coded data at the encoder. The proposed methods n [1]- [3] are among the methods n the frst group. In [1] the nformaton provded by the ER tools n MPEG-4 coded vdeo streams are further processed to detect the transmsson errors. [2] ntroduces a method for the erroneous macro blocks n the decoded frames of H.263 bt stream by checkng a set of error detecton condtons derved from the redundant nformaton nherent wth n the frame. In [3] the non-exhaustve reversble varable length codng (RVLC) s used to detect the errors n the coded MPEG-4 and H.263 bt streams.. The proposed method n [4] employs an nserted fragle watermark n the encoder sde for error detecton and localzaton n wreless vdeo communcatons. Turbo codes are also used for error detecton n [5]. The Reed-Solomon codes are used n [6] for a fully JPEG 2000 complant backward compatble error protecton for headers. In [7] the codng redundancy s ntroduced by means of a modfed context based arthmetc entropy coder (CBAEC), whch enhances the error reslence of the JPEG There are seven ER tools n the JPEG2000 standard Part 1 [8] that are used n two dfferent levels. The frst fve error reslence tools are used at the entropy codng level and the other two are used at the packet level. The error reslence tool n JP2 that s used for error detecton s the segmentaton symbol. It s used for detectng bt errors that may occur n the CBAEC data durng transmsson or storage. Segmentaton symbol s a specal symbol whch s coded at the end of each bt-plane. In spte of the common belef that correct decodng of ths symbol ensures error free decodng of the bt-plane [8], there mght be false postves for ncorrect bt-planes among the bt-planes wth correctly decoded segmentaton symbols. The undetected ncorrect bt-planes may result n annoyng artfacts n the decoded mages. In ths work we ntroduce a novel error concealment method that can be used for detecton and elmnaton of the artfacts caused by the damaged btplanes, not detected by the segmentaton symbols. The proposed method s mplemented n a JPEG2000 codec and through extensve smulatons t s shown that the proposed method can mprove the error reslency of the JPEG2000 codec by up to 6 db, dependng on the poston of error and mage content. The paper s organzed as follows. In secton 2, we provde a bref overvew of JPEG2000 compresson and dscuss the problem of false postve segmentaton symbol. The proposed codec s dscussed n secton 3. Smulatons results are

2 provded n secton 4 and the concludng remarks are gven n secton 5. II. BACKGROUND In the JPEG2000 standard, mage compresson s carred out at several steps. Frst, an mage s dvded nto non overlapped tles. Then, n order to reduce the redundances among the color components, a component transform s used for each tle. Dscrete wavelet transform s appled to each resulted color component to generate the wavelet-coeffcents at varous resoluton levels and four LL, HL, LH and HH subbands. The resulted coeffcents are dvded nto non overlapped blocks named code blocks. Each code block s coded ndependently, one bt-plane at a tme. A context based arthmetc entropy coder (CBAEC) encodes each bt-plane n three passes. Detaled descrpton of JPEG2000 mage codng standard can be found n [8]-[10]. CBAEC has hgher compresson effcency than other entropy codng methods such as Huffman codng. However, t s more senstve to transmsson bt errors, because t s mpossble to locate the bt errors n the CBAEC coded sequence. Hence, f a bt error occurs wthn a CBAEC coded stream, the whole coded sequence should be dscarded. The JPEG2000 standard has ntroduced a segmentaton symbol for detectng the damaged CBAEC coded bt-planes. The Segmentaton symbol s a symbol 0xA or 1010 whch s coded by CBAEC at the end of each bt-plane. In a code block, f the decoder does not decode the segmentaton symbol for the n th bt-plane, the n th bt-plane s consdered as damaged and ths bt-plane as well as the followng bt-planes n the same code block wll be dscarded. However, n the same code block, the (n-1) th bt-plane may also be damaged, even f the segmentaton symbol for ths bt-plane s decoded correctly. Each bt n the segmentaton symbol s coded by the unform context wth a probablty close to 0.5. Hence, the probablty that the segmentaton symbol n a damaged bt-plane s decoded correctly s (1/2) 4 = = 1/16. It means that, the (n-1) th bt-plane of a code block wth damaged n th bt-plane may also be a damaged bt-plane wth the probablty of 1/16. Hence both (n-1) th and (n-2) th bt-planes n the same code block can be damaged wth a probablty of 1/16 1/16 = 1/256. In practce the latter case due to ts very low probablty can be gnored. However, the frst case ndcates that one code block out of 16 damaged code blocks may have an erroneous bt-plane, whle ts segmentaton symbol s correctly decoded. Fg. 1 shows an example of the artfacts produced by such corrupted bt-plane. Note that a damaged bt-plane may have a large number of ncorrectly decoded bts and usng t for reconstructon of the receved mage, results n consderable artfacts. Moreover, the damaged bt-planes that are not detected by the segmentaton symbol are consdered as correct data and may be used at the error concealment stage for the recovery of other damaged data. The error concealment n ths case causes propagaton of error to the recovered data. Hence, detectng and removng the damaged bt planes by the Fg. 1. An example of artfacts caused by a hghly damaged btplane n a JP2 mage. proposed method wll also mprove the performance of error concealment. Even though error detecton methods for the JP2 coded mages have appeared n the lterature [6] [7], to the authors best knowledge there s not any method for detectng such damaged bt-planes n a sngle JP2 coded mage. Detectng ncorrect bt-planes among the bt-planes wth the correctly decoded segmentaton symbol n Moton JPEG2000 coded mage sequences s dscussed n [11]. They have used the temporal redundances among the consecutve mages of Moton JPEG2000 coded mage sequences for detectng and removng the hghly damaged bt-planes wth false postve segmentaton symbol. Snce the proposed method n I [11] s not applcable for a sngle JP2 coded mage, n ths paper we propose an error detecton method for detectng hghly damaged bt-planes, whch have not been detected by the segmentaton symbol n a sngle coded mage. The proposed method s based on modelng the coded bt-plane at the encoder and usng ths model at the decoder to mprove the detecton of the errors n the bt-planes of a sngle JP2 coded mage. We have tested the proposed method on a JP2 coder and extensve smulatons ndcate that the proposed method can remove the artfacts caused by the hghly damaged btplanes and mproves the qualty of decoded mages by up to 6 db n PSNR. III. THE PROPOSED ERROR DETECTION METHOD The mpact of a damaged bt-plane on the qualty of a decoded mage depends on the locaton of the bt error. If the bt error occurs at the begnnng of a coded bt-plane, a large number of decoded bts n the damaged bt-plane wll be affected and they may take wrong values. In ths case, the damaged bt-plane causes artfacts n the form of noses n a rectangular regon of the decoded mage, smlar to those n

3 cancelng only the hghly damaged bt-plane n code block 15 mproves the PSNR a)psnr=31.2 db b)psnr=31.2 db Fg. 3. Error concealed pcture of Fg 1 after removng, a) only damaged block 15 and b) all the 5 damaged blocks. Fg. 2. The Wavelet transform of Fg. 1. Squares show the damaged code blocks. The code block number 15 s the hghly damaged code block, causng vsble artfacts n Fg. 1. Fg. 1. If the decoder can detect these hghly damaged btplanes n the decodng process and elmnate them, the decoded mage wll be free of artfacts and ts qualty wll be mproved sgnfcantly. In ths paper, we propose a new method for detectng the damaged bt-planes that are not detected by the segmentaton symbol. The proposed method s explaned n the followng. Values of the decoded bts n a bt plane are ether 0 or 1 but the number of bts wth a value equal to 0 s much larger than the number of bts wth a value equal to 1. Ths s because wavelet s a sparse transform and most of the wavelet-coeffcents have small values close to zero. Ths s partcularly true at the earler passes of codng. When, due to channel error, a large number of bts n a damaged bt-plane are changed, there wll be many bts wth the ntal value of 0 that are decoded as 1. In ths case, the number of decoded bts wth values equal to 1 wll change drastcally. As an example, Fg. 2 shows the dscrete wavelet transform of Fg. 1. For better vsualzaton of the hgh frequency wavelet coeffcents, they are magnfed by a factor of 8 and added to +128 (except the lowest LL band). The damaged code block 15 n the upper left sde of Fg. 2 s the hghly damaged bt plane, whch resulted n the artfacts n Fg. 1. As Fg. 2 ndcates, there are a large number of whte pxels n the code block 15. These whte pxels mostly come from the bts n a damaged bt-plane wth value equal to 0 that are decoded as 1, due to bt error n the compressed mage. The hghly damaged code block 15 n Fg. 2 shows that n a hghly damaged bt-plane the number of decoded 1s has ncreased consderably. It s worth notng that n Fg. 2, the code blocks 61, 102, 104 and 241 have undetected damaged bt planes as well, but ther artfacts are not as vsble as block 15. Whle the PSNR of the mage n Fg. 1 s 27.6 db to 31.2 db, as shown n Fg 3a. Had we concealed all the 5 damaged blocks, the PSNR s agan 31.2, as shown n Fg 3b. These ndcate that elmnatng only the hghly damaged btplane removes the man artfact from the decoded mage and the PSNR mprovement n ths case s comparable wth the case where all the damaged code blocks have been detected and elmnated. Snce the number of decoded 1s s the key pont n detecton of the hghly damaged bt-planes, we have ntroduced a method for estmatng the number of decoded 1 s for each bt-plane. The necessary nformaton for estmatng the number of 1s s sent to the decoder and the decoder uses ths nformaton to decde f a bt-plane s damaged. At each bt plane, dependng on the bt-plane number, the number of decoded 1s s estmated by two technques. One s used for all the bt-planes except the most sgnfcant bt (MSB) and the other for the MSB. For non MSB, the number of 1s n a btplane s modeled usng parameters from the current bt plane and the prevously coded bt-plane. We use the lnear Equaton (1) as an expermental model for estmatng the number of decoded bts wth values equal to 1 n the coded bt-plane: O = a1 L + a2o 1 + a3l 1 + a4 + a5r (1) where O and O 1 are the normalzed number of 1s n the bt-plane number and 1 respectvely, L and L 1 are the normalzed length of the coded bt stream for bt-planes and 1respectvely, R s the resoluton level number and s the bt-plane number. We dvde the number of 1s n the btplane to the total number of wavelet-coeffcents of a code block to normalze the number of 1s n the bt-plane and the normalzaton of the length of the coded bt stream for a btplane s done n the same way. At the encoder, we also use the least square error (LSE) curve fttng technque to estmate the coeffcents a 1 to a 5 n the model of equaton (1). Ths model provdes an estmate for the number of 1s n each btplane of the encoded mage (except for the frst bt-plane). Moreover, the boundares of the relatve errors n the estmaton of the number of 1s n the above model ( Emn and E max ) are determned at the encoder.

4 The encoder sends the model coeffcents ( a1 to a 5 ) and also the relatve error boundares ( Emn and E max ) along wth the coded mage. For each decoded code block, the decoder compares the normalzed number of 1s n the last correctly decoded bt-plane ( O' ) wth the predcted value ( OE ) by the model. If t does not satsfy the followng nequalty: O' OE Emn Emax (2) OE then, the bt-plane n the correspondng code block wll be consdered as a damaged bt-plane and all the bts n ths btplane wll be set to zero. Ths model cannot be appled to the most sgnfcant btplane n the code block because t uses the prevous bt-plane nformaton whch s of course not avalable for the most sgnfcant bt-plane. Detectng the damaged bt-planes among the most sgnfcant bt-planes s based on the fact that there are only a few coeffcents wth non-zero values n the most sgnfcant bt-plane of the code blocks n the HL, LH and HH sub-bands. In ths case the maxmum value of the normalzed number of 1s n the most sgnfcant bt-plane among all the code blocks n HL, LH and HH sub-bands ( O 1 MAX ) s calculated and s sent to the decoder. When the decoder receves only the frst bt-plane wth the correct segmentaton symbol n a code block, t calculateso ' 1 for the receved frst bt-plane. If the calculatedo ' 1 for ths bt-plane s hgher than the receved maxmum value for O 1 ( O 1 MAX ), the decoded bt-plane s consdered as damaged and all of the bts n the detected damaged bt-plane wll be set to zero. We note that the modelng equaton can be put n the JPEG2000 COMMENT marker segment so that the JPEG2000 code stream syntax s not volated. In the proposed method, the transmtted coeffcents nclude sx floatng values for coeffcents a 1 to a 5 ando 1 MAX, and two shortnteger values for E mn and E max. Hence, the total number of extra bytes wll be =28 bytes, whch s a neglgble overhead compared to the compressed mage data rate. Moreover, Forward Error Correcton (FEC) can be used to ensure secure transmsson of the 28 extra bytes. IV. SIMULATIONS RESULTS Snce the proposed model s an expermental model, extensve smulatons were run to verfy ts performance. The proposed method was mplemented n a JP2 codec and tested wth three groups of mages. The frst group ncluded standard gray scale mages such as Barbara, Goldhll, Lena, and Peppers and the second group conssted of sngle frames of QCIF ( ) standard mage sequences such as Carphone, Salesman, Mother-Daughter and Foreman. The thrd group ncluded 1235 mages from the Benchathlon [12] mage database wth resolutons and ,converted to 8 bt gray scale. Each mage was encoded wth a JP2 encoder. The codng was lossless wth 5 levels of 5/3 wavelet transform and code block sze. The segmentaton symbols were also ncluded n the coded mages. The modelng parameters were calculated n the codng process for each mage and saved n the comment marker segment of the coded mage. A channel error smulator nserted random bt errors n the data secton of the coded JP2 fles. We made 100 damaged mages from each test mage n the frst group, usng the channel error smulator wth 1E-3 and 1E-4 bt error rates (BER). As mentoned, the number of undetected damaged bt planes s about 1/16 of the total number of damaged bt planes and for a fxed BER the number of damaged bt planes s proportonal to both the mage sze and the number of code blocks n the coded mage. Ths means for mages n the second group of the test, that have a much lower resoluton than the mages n the frst group and are coded wth equal code block sze, there s much less probablty that for the same BER a damaged bt plane n the coded mage to be undetected. Hence, we made 500 damaged mages from each test mage n the second group, usng the channel error smulator wth 1E-3 and 1E-4 BER. In the thrd group for each mage we made 10 damaged samples wth BER equal to 1E-3 and another 10 damaged samples wth BER of 1E-4. There were on average about two samples wth artfacts due to the false postve segmentaton symbols among the 10 damaged samples wth a BER of 1E-3 and one sample wth smlar artfacts among the 10 damaged samples wth a BER of 1E-4. The damaged JP2 encoded mages were decoded wth three dfferent decoders. The frst decoder, D1, was the reference JP2 decoder. Ths decoder dscarded a decoded btplane wth an ncorrect segmentaton symbol along wth ts followng bt-planes. The second decoder, D2, was the deal error detecton decoder and s referred to as the deal decoder hereafter. The deal decoder practcally s not realzable, because t requres the orgnal mage to detect the errors. Ths decoder compares the bt-plane located just before a damaged bt-plane, wth the correspondng bt-plane n the orgnal error free encoded mage. If there was any dfference between the compared bt-planes, the deal decoder elmnates the tested bt-plane besdes the detected damaged bt-planes. Hence, ths deal decoder detects and removes all the damaged bt-planes that have not been detected by the segmentaton symbol. Snce n a sngle JP2 mage there s no other algorthm for detectng the damaged bt-planes that are not detected by the segmentaton symbol, we have evaluated the performance of the proposed method by comparng t wth an deal decoder. The last decoder, D3, was a decoder whch apples the proposed error detecton method. Decoder D3 uses the embedded modelng parameters n the comment marker segment of the receved damaged mage to estmate the number of 1 s n the bt-plane just before a detected damaged bt-plane. Decoder D3 compares the number of decoded 1s n the tested bt-plane wth the estmated value and f t falls out of the permssble error boundares, the decoder elmnates the tested bt-plane n addton to the detected damaged bt-planes.

5 a) PSNR=26.3 db b) PSNR=32.3 db c) PSNR=32.3 db d) PSNR=26.9 db e) PSNR=32 db f) PSNR=30.1 db g) PSNR=17.14 db h)psnr=18.37 db )PSNR=17.95 db Fg. 4. The decoded mages usng decoders D1 (left column), D2 (mddle column) and D3 (rght column) Fg. 4 shows some of the decoded mages by these three decoders n the three test mages groups. As shown n Fg. 4, the artfacts that are n the decoded mages of decoder D1 are completely removed n the mages decoded by an deal decoder, D2, and our decoder D3. Moreover, the performance of the proposed method n decoder D3 s very close to the deal decoder D2, both subjectvely and also n terms of objectve measures. The overall performance mprovement of our decoder over the conventonal decoder, D3-D1 compared to that of the deal decoder D2-D1 for the frst set of mages s shown n Table 1.A smlar comparson wth the other two set of mages are respectvely shown n Tables 2 and 3. Tables 1-3 ndcate that our proposed decoders D3 lke the deal decoder D2, outperforms the conventonal decoder of the JPEG2000, D1. Moreover, we can deduce from Tables 1-3 that the performance of the proposed algorthm s smlar to the deal decoder and n some cases even outperforms the deal decoder. As we mentoned at the begnnng of secton 3, the mpact of a damaged bt-plane on the qualty of a decoded mage, depends on the locaton of the bt error. If an error occurs at the end of a bt-plane, the decoded bt-plane wll have a large number of correctly decoded bts and few bts wth ncorrect values. It s obvous that elmnatng such a bt plane wll reduce the mage qualty and also the PSNR of the decoded mage. Ths s why the deal decoder that removes all the damaged bt-planes, ncludng the hghly damaged bt-planes and the bt-planes wth a few damaged bts, n some cases has an nferor performance to the proposed decoder, whch elmnates only the hghly damaged bt-planes. Snce the bt error has a random characterstc, the resulted artfacts vary n dfferent damaged mages, and so does the resultant mprovement, when the artfacts are removed usng the proposed method. In our smulatons, the maxmum acheved mprovement for the recovered mages usng the proposed error concealment method was 6

6 db n PSNR (Fg. 4.c), and about 1.2 db on average for the PSNR of total number of recovered mages wth artfact. It s worth notng that, as the recovered mages ndcate, the subjectve qualty of the decoded mages s hgher than what s measured by the PSNR. V. CONCLUSIONS We have proposed a new error detecton method for the JPEG2000 (JP2) mage codng standard. The proposed error detecton method enables the decoder to remove the artfacts n the damaged JP2 coded mages caused by the damaged bt-planes that were not detected by the segmentaton symbol. In our approach, we provde a model for estmatng the value of the number of non-zero bts n an encoded bt-plane at the encoder sde. The model conssts of eght parameters, whch only requre 28 bytes overhead. The modelng parameters are accommodated n the comment marker segment of the JPEG2000 code stream BER 1E-4 1E-3 BER 1E-4 1E-3 BER TABLE I RELATIVE IMPROVEMENT IN PICTURE QUALITY WITH THE FIRST SET OF IMAGES Image name No. of mages wth artfacts n 100 damaged mages to mantan the JPEG2000 bt stream syntax. At the decoder, the bt-plane just before a damaged bt-plane and wth the number of non-zero bts hgher than the upper estmated value by the transmtted model wll be consdered as a damaged bt-plane too. Extensve smulatons results over a large number of mages and wde range of BERs show that the proposed method successfully removes the artfacts caused by the damaged bt-planes, whch have not been detected by the conventonal segmentaton symbol. Snce the mposed error and ts mpacts on the decoded mages have random characterstcs, the resulted PSNR of varous decoded erroneous mages vares. However, mprovements up to 6 db n PSNR have been acheved n our smulatons and the average acheved mprovement n PSNR by removng the artfacts n the damaged mages was about 1.2 db. artfact usng D2 artfact usng D3 Barbara GoldHll Lena Peppers Barbara GoldHll Lena Peppers TABLE II RELATIVE IMPROVEMENT IN PICTURE QUALITY WITH THE SECOND SET OF IMAGES Image name No. of mages wth artfacts n 500 damaged mages artfact usng D2 artfact usng D3 Carphone Foreman Foreman Mthr_dotr Salesman Carphone Foreman Foreman Mthr_dotr Salesman TABLE III RELATIVE IMPROVEMENT IN PICTURE QUALITY WITH THE THIRD SET OF IMAGES Average No. of mages wth artfacts n 10 damaged mages (Total No.generated damaged mages was for each BER) artfact usng D2 compared to D1 decoded mages wth artfact usng D3 1E E

7 REFERENCES [1] Bo Yan and Kam Wng Ng, Analyss and Detecton of MPEG-4 Vsual Transmsson Errors over Error-Prone Channels, IEEE Transactons on Consumer Electroncs, Vol. 49, No. 4, pp , NOVEMBER [2] E. Khan, S. Lehmann, H. Gunj, and M. Ghanbar, Iteratve error detecton and correcton of H.263 coded vdeo for wreless networks, IEEE Transactons on Crcuts and Systems for Vdeo Technology, VOL. 14, No. 12, pp , DECEMBER [3] Cha-We Ln and Ja-Lng Wu, On Error Detecton and Error Synchronzaton of Reversble Varable-Length Codes, IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 53, NO. 5, pp , MAY [4] M. Chen, Y He, R. L. Lagendjk, A Fragle Watermark Error Detecton Scheme for Wreless Vdeo Communcatons, IEEE Transactons on Multmeda, VOL. 7, No. 2, pp , APRIL [5] Fengqn Zha, and Ivan J. Far, Technques for Early Stoppng and Error Detecton n Turbo Decodng IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 51, NO. 10, pp , OCTOBER [6] Dder Ncholson, Catherne Lamy-Bergot, Xaver Naturel and Charly Poullat, JPEG 2000 Backward Compatble Error Protecton wth Reed- Solomon Codes, IEEE Transactons on Consumer Electroncs, Vol. 49, No. 4, pp , NOVEMBERMBER [7] Marco Grangetto, Enrco Magl, and Gabrella Olmo, A Syntax- Preservng Error Reslence Tool for JPEG 2000 Based on Error Correctng Arthmetc Codng, IEEE TRANSACTIONS ON IMAGE PROCESSING, VOL. 15, NO. 4, pp , APRIL [8] ITU-T Recommendaton T.800, JPEG2000 mage codng system: Core codng system, August [9] D. Taubman, and M.W. Marcelln, JPEG2000 Image Compresson Fundamentals, Standards and Practce, Kluwer Academc Publshers, Boston/Dordrecht/London, [10] T. Acharya, P. S. Tsa, JPEG2000 Standard for Image Compresson Concepts, Algorthms and VLSI Archtectures, John Wley & Sons Inc., Hoboken, New Jersey, [11] Farzad Zargar, Mahmoud Reza Hashem, An Improved Moton JPEG2000 Decoder for Error Concealment of Segmentaton Symbol Faults, Accepted paper n EURASIP Journal on Advances n Sgnal Processng [12] /mg/done/ bomedcal engneerng, multmeda nformaton codng, securty and retreval, and human-computer nteracton systems. Mohammad Ghanbar (M 78-SM 97-F01) s a Professor of Vdeo Networkng n the Department of Computng and Electronc Systems, Unversty of Essex, Unted Kngdom. He s best known for the poneerng work on two-layer vdeo codng for ATM networks, now s known as SNR scalablty n the standard vdeo codecs, whch earned hm the Fellowshp of IEEE n He has regstered for eleven nternatonal patents and publshed more than 400 techncal papers on varous aspects of vdeo networkng. He was the corecpent of A.H. Reeves prze for the best paper publshed n the 1995 proceedngs of IEE n the theme of dgtal codng. He was also a convestgator of the European MOSAIC project studyng the subjectve assessment of pcture qualty, whch resulted to ITU-R Recommendaton 500. He s the co-author of Prncples of Performance Engneerng, book publshed by IEE press n 1997, and the author of Vdeo codng: an ntroducton to standard codecs, book also publshed by IEE press n 1999, whch receved the Raylegh prze as the best book of year 2000 by IEE. Hs latest book Standard Codecs: mage compresson to advanced vdeo codng, was publshed by IEE n He has been an organzng member of several nternatonal conferences and workshops. He was the general char of 1997 nternatonal workshop on Packet Vdeo and Guest Edtor to 1997 IEEE Transactons on crcuts and systems for Vdeo Technology, Specal ssue on Multmeda technology and applcatons. He has served as Assocate Edtor to IEEE Transactons on Multmeda (IEEE-T-MM from ) and represented Unversty of Essex as one of the sx UK academc partners n the Vrtual Centre of Excellence n Dgtal Broadcastng and Multmeda. He s a Fellow of IEEE, Fellow of IET and Charted Engneer (CEng). Farzad Zargar receved hs B.Sc. degree n Electrcal Engneerng from Sharf Unversty of Technology and hs M.Sc. and Ph.D. degrees n Electrcal Engneerng from Unversty of Tehran, all n Tehran, Iran. He s currently a research assocate at the nformaton technology research nsttute of Iran Telecom Research Center (ITRC), Mnstry of Telecommuncatons and Informaton Technology of Iran. Hs research nterests nclude mage and vdeo sgnal processng algorthms and multmeda systems. Al Sefdpour receved hs B.Sc. and M.Sc. degrees n computer engneerng from Iran Islamc Azad Unversty, Tehran Center Campus, Tehran, Iran. He s currently the M.Sc. student n the computer engneerng department of Iran Islamc Azad Unversty, Scence and Research Campus, Tehran, Iran. Hs research nterests nclude mage codng and encrypton. M. Shahram Mon receved hs B.Sc. degree n electrcal engneerng from Amr Kabr Unversty of Technology, and M.Sc. degree n electrcal engneerng from Tehran Unversty, both n Tehran, Iran and Ph.D. degree n electrcal engneerng from École Polytechnque de Montréal, Montréal, Canada. He s currently the drector of the Multmeda Research Group at Iran Telecom Research Center (ITRC), Mnstry of Telecommuncatons and Informaton Technology of Iran. Hs research nterests nclude applcatons of mage processng, and pattern recognton algorthms n