`
`
`
`
`
`
`
`
`
`
`
`IMAGE
`
`
`
`
`
`
`
`
`
`
`
`
`COMMUNICATION
`
`—
`
`Papers
`
`CONTENTS
`
`J.T. Virtamo and S.T, Valli
`Vector quantization with hierarchical classification of sub-blocks..................005
`G. Keesman, R. Hellinghuizen, F. Hoeksema and G. Heideman
`"FRAnsCadi ne OL MEber BitSitdaise nce acane-s ‘aecest
`aoe nwo Ge an» 80a hoe whe pom bia Sao
`
`
`
`Y.K. Kim and J.B. Ra
`
`Multiple shell structured hypercube feature maps for vector quantization of images. .......
`L. Wu, J. Benois-Pineau, P. Delagnes and D. Barba
`
`Spatio-temporal segmentation of image sequencesfor object-oriented low bit-rate image coding
`M.K.N. Schéyer and P.W. Verbeek
`Block position dithering in DCT-coded sequences... .. 2... 2-5. cee ee ee tes
`
`
`
`
`
`P, Franti, T. Kaukoranta and O. Nevalainen
`
`
`
`
`
`
`
`
`On the design of a hierarchical BTC-VQ compression system. .........--.-+000+0e%
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`PROG AE8 8 i.e td bee MR I ree eke nv:
`in eet esis nde Apo:
`@ onto
`
`
`
`
`
`Cunitilatve cantante oF Valiime 8 ec se Ve ae bce Be a RR Ges ek
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`LAST NUMBER OF THIS VOLUME
`
`lechniques & Applications
`Theory.
`A publicationof the European Association for Signal Processing (EURASIP) rij
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`Page 1 of 23
`
`GOOGLE EXHIBIT 1005
`
`

`

`SIGNAL PROCESSING
`
`
`
`
`
`
`IMAGE COMMUNICATION
`
`
`
`Theory, Techniques & Applications
`
`
`
`
`A publication of the European Association for Signal Processing (EURASIP)
`
`
`
`
`
`
`
`
`
`
`
`Editor-in-Chief
`
`Leonardo CHIARIGLIONE
`
`
`Centro Studi e Laboratori
`
`
`
`
`Telecommunicazioni (CSELT)
`
`
`
`Via Guglielmo Reiss-Romoli, 274
`
`
`
`
`1-10148 Torino, Italy
`
`
`
`
`Telephone: (11) 2286 120
`
`
`
`Telefax: (11) 2286 299
`
`
`
`
`
`
`
`Editorial Board
`
`
`J. Biemond (Univ. Delft, The Netherlands)
`
`
`
`
`
`G. Boerger (HHI, Germany)
`
`
`
`
`E. Dubois (Univ. Quebec, Canada)
`
`
`
`
`B. Girod (Erlangen, Germany)
`
`
`
`
`
`
`
`
`
`
`
`Guest Editors
`
`
`P, Aigrain (IRIT, France)
`
`
`
`
`S.A. Benton (MIT, USA)
`
`
`
`
`G
`De Haan (Philips, The Netherlands)
`
`
`
`
`
`Fujii (NTT, Japan)
`
`
`
`
`Guichard (CNET, France)
`
`
`
`
`Hamasaki (Univ. Tokyo, Japan)
`
`
`
`
`
`Johann (Deutsche Bundespost, Germany)
`
`
`
`
`Kurita (NHK. Japan)
`
`
`
`
`LeGall (C-Cube, USA)
`
`
`
`
`Liu (Princeton Univ., USA)
`
`
`
`
`
`Morrison (BT Labs, United Kingdom)
`
`
`
`
`
`Narasimhalu (Singapore)
`
`
`
`Nicol (BT Laboratories)
`
`
`
`
`. Okubo (NTT, Japan)
`
`
`
`
`. Omachi (NEC, Japan)
`
`
`
`
`A.P. Pentland (MIT, USA)
`
`
`
`M. Shibata (NHK, Japan)
`
`
`
`
`T. Sikora (HHI, Berlin)
`
`
`
`
`W. Verbiest (Belgium)
`
`
`
`E. Viscito (Chromatic Research Corporation, USA)
`
`
`
`
`
`
`
`
`C.N. Judice (Bell, USA)
`
`
`
`
`J.-K. Kim (KAIST, South Korea)
`
`
`
`
`
`A.B. Lippman (MIT, USA)
`
`
`
`
`
`H.-G. Musmann (Univ. Hannover, Germany)
`
`
`
`
`
`D, Nasse (CCETT, France)
`
`
`
`
`¥. Ninomiya (NHK, Japan)
`
`
`
`
`D. Pearson (Univ, Essex, United Kingdom)
`
`
`
`
`
`
`F. Pereira (IST, Portugal)
`
`
`
`
`L. Stenger (Fl DBP, Germany)
`
`
`
`
`
`H. Tominaga (Waseda Univ., Japan)
`
`
`
`
`M. Tanimoto (Nagoya Univ., Japan)
`
`
`
`
`
`M. Vetterli (Lausanne, Switzerland)
`
`
`
`
`L.T. Wu (ITRI, Taiwan)
`
`
`
`
`H. Yasuda (NTT, Japan)
`
`
`
`
`
`
`
`
`aece4
`O000
`AwD
`
`
`
`
`
`
`
`
`
`IMAGE COMMUNI-
`Editorial Policy. SIGNAL PROCESSING:
`
`
`
`
`
`
`CATIONis an international journal for the development of the
`
`
`
`
`
`
`
`
`
`
`theory and practice of image communication. Its primary objec-
`
`
`
`
`
`
`
`
`
`tives are the following:
`
`
`
`
`To present a forum for the advancement of the theory and
`
`
`
`
`
`
`
`
`
`
`
`practice of image communication.
`
`
`
`
`To stimulate cross-fertilization between areas similar in nature
`
`
`
`
`
`
`
`
`which havetraditionally been separated, for example, various
`
`
`
`
`
`
`
`
`aspects of visual communications and information systems.
`
`
`
`
`
`
`
`- To contribute to a rapid information exchange between the
`
`
`
`
`
`
`
`
`
`industrial and academic environments.
`
`
`
`
`Theeditorial policy and the technical contentof the journal are the
`
`
`
`
`
`
`
`
`
`
`
`
`responsibility of the Editor-in-Chief and the Editorial Board. The
`
`
`
`
`
`
`
`
`
`Journalis self-supporting from subscription income and contains
`
`
`
`
`
`
`
`
`a minimum amountof advertisements. Advertisements are subject
`
`
`
`
`
`
`
`
`to the prior approval of the Editor-in-Chief. The journal welcomes
`
`
`
`
`
`
`
`
`
`
`contributions from every country in the world.
`
`
`
`
`
`
`
`
`IMAGE COMMUNICATION
`Scope. SIGNAL PROCESSING:
`
`
`
`
`publishesarticles relating to aspects of the design, implementation
`
`
`
`
`
`
`
`
`and use of image communication systems.
`
`
`
`
`
`
`SIGNAL PROCESSING:
`IMAGE COMMUNICATION features
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`original research work, tutorial and review articles, and accountsof
`practical developments.
`
`
`
`
`
`
`Subjects. Subject areas covered by the journal include:
`
`
`
`
`
`
`
`
`TV, HDTV and 3D-TV systems
`Image Transmission
`
`
`
`
`
`
`
`Visual Science
`interactive Image
`
`
`
`
`Image Coding
`Communication
`
`
`
`
`
`
`TV and Advanced TV
`
`
`
`Broadcasting
`
`Image Storage and Retrieval
`
`
`
`Graphic Arts
`
`
`Electronic Printing
`
`
`
`
`
`
`Imaging Technology
`
`Display Technology
`
`
`VLSI Processors for
`
`
`
`Image Communications
`
`
`
`
`
`
`
`
`
`Membership and Subscription Information.
`
`
`
`
`SIGNAL PROCESSING: IMAGE COMMUNICATION(ISBN 0923-
`
`
`
`
`
`5965)is published in one volume (six issues) a year. For 1996
`
`
`
`
`
`
`
`
`
`
`
`Volume 8 is scheduled for publication. Subscription prices are
`
`
`
`
`
`
`
`
`available upon request from the publisher. Subscriptions are ac-
`
`
`
`
`
`
`
`
`cepted on a prepaid basis only and are entered on a calendar year
`
`
`
`
`
`
`
`
`
`
`
`basis. Issues are sent by surface mail except to the following
`
`
`
`
`
`
`
`
`
`
`
`countries, where air delivery (S.A.L.
`~ Surface Air Lifted) is en-
`
`
`
`
`
`
`
`
`
`
`sured: Argentina, Australia, Brazil, Canada, Hong Kong, India,
`
`
`
`
`
`
`
`
`israel, Japan, Malaysia, Mexico, New Zealand, Pakistan, People’s
`
`
`
`
`
`
`
`
`Republic of China, Singapore, South Africa, South Korea, Taiwan,
`
`
`
`
`
`
`
`
`
`Thailand, USA, For the rest of the world, airmail charges are
`
`
`
`
`
`
`
`
`
`
`
`available upon request. Claims for missing issueswill be honoured
`
`
`
`
`
`
`
`
`
`
`free of chargeif made within six months after the publication date
`
`
`
`
`
`
`
`
`
`
`
`of the issues. Mail orders and inquiries to: Elsevier Science B.V.,
`
`
`
`
`
`
`
`
`
`
`
`Journals Department, P.O. Box 211, 1000 AE Amsterdam, The
`
`
`
`
`
`
`
`
`
`
`Netherlands. Forfull membership information of the Association,
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`possibly combined with a subscription at a reduced rate, please
`contact EURASIP, P.O. Box 134, CH-1000 Lausanne 13,
`
`
`
`
`
`
`
`
`
`Switzerland.
`
`
`
`
`'
`1996 Elsevier Science B.V. All rights reserved
`
`
`
`
`
`
`Nopart of this publication may be reproduced, stored in a retrieval system or transmitted in any form or by any means,electronic, mechanical, photocopying, recording or
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`otherwise, without the prior permission of the publisher, Elsevier Science B.V., Copyright and Permissions Department, P.O. Box 521, 1000 AM Amsterdam, The Netherlands
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`Special regulations for authors. Upon acceptance of an article by the journal. the author(s) will be asked to transfer copyrightof the articie to the publisher. This transfer will
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`ensure the widest possible disseminationof information.
`
`
`
`
`
`
`
`Special regulations tor readers in the USA. This journal has been registered with the Copyright Clearance Center, Inc Consentis given for copying ofarticles for personal or
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`internal use, or for the personaluse of specific clients. This consent is given on the condition that the copier pays through the Center the per-copy fee stated in the code on the
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`first page of each article tor copying beyand that permitted by Sections 107 or 108 of the US Copyright Law. The appropriate fee should be forwarded with a copyof thefirst page
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`of the article to the Copynght Clearance Center, Inc., 222 Rosewood Drive, Danvers, MA 09123, USA.If no code appears in anarticle, the author has not given broad consentto
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`copy and permission to copy must be obtained directly trom the author. The fees indicated on the first page of an article in this issue will apply retroactively to ail articles
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`published in the journal, regardless of the yearof publication. This consent does not extend to other kinds of copying, such as for generaldistribution, resale, advertising and
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`promotion purposes, or for creating new collective works. Special written permission must be obtained from the publisher for such copying,
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`Noresponsibility is assumed by the Publisher for any injury and/or damage to persons or property as a matterof productsliability, negligence or otherwise, or from any use oF
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`operation of any methods, products, instructions or ideas contained in the material herein.
`
`
`
`
`
`
`
`
`
`
`
`
`
`Although all advertising material is expected to conform to ethical standards, inclusion in this publication does not constitute a guarantee or endorsementof the quality or value
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`of such productor of the claims made ofit by its manufacturer.
`
`
`
`
`
`
`
`
`
`
`
`
`68 The paperusedin this publication meets the requirements of ANSI/NISO 239.48-1992 (Permanence of Paper).
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`Published 6 times a year
`
`
`
`
`
`
`0923-5965 /96/$15.00
`
`
`
`
`
`
`
`Printed in The Netherlands
`
`
`
`Page 2 of 23
`
`Page 2 of 23
`
`

`

`Volume 8, No. 6, September 1996
`
`CONTENTS
`
`Papers
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`475
`
`481
`
`501
`
`513
`
`545
`
`
`551
`
`563
`565
`
`
`
`
`
`
`
`
`
`
`
`http://www.elsevier.nifocate/eee.
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`J.T. Virtamo and §S.T. Valli
`Vector quantization with hierarchical classification of sub-blocks.............
`G. Keesman, R. Hellinghuizen, F. Hoeksema and G. Heideman
`
`Transcoding of MPEG bitstreams... ........-0 00.0. eee ee eee eee
`Y.K. Kim and J.B. Ra
`Multiple shell structured hypercube feature maps for vector quantization of images.
`
`L. Wu, J. Benois-Pineau, P. Delagnes and D. Barba
`Spatio-temporal segmentation of image sequences for object-oriented low bit-rate
`
`
`
`
`
`image coding... 1.0...ee ee ee
`
`
`
`
`
`
`
`M.K.N. Schéyer and P.W. Verbeek
`
`
`
`
`
`
`
`
`
`Block position dithering in DCT-coded sequences... . 1... 0.0.2... ee eee
`
`
`
`
`P. Franti, T. Kaukoranta and O. Nevalainen
`
`
`
`
`
`On the design of a hierarchical BTC-VQ compression system...........-.-.-
`
`
`
`
`
`
`
`
`
`
`
`Author index of Volume 8... 0.0.0.0... 0.00000 ce ee ee eee
`
`
`
`
`
`
`
`
`
`Cumulative contents of Volume 8... .......0.0... 0002 eee eee eee eens
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`Advancetables of contents can be retrieved at
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`Page 3 of 23
`
`Page 3 of 23
`
`

`

`
`
`IMAGE
`
`COMMUNICATION
`
`
`Transcoding of MPEG bitstreams
`
`
`
`Gertjan Keesman
`
`
`
`
`Robert Hellinghuizen®, Fokke Hoeksema”, Geert Heideman?
`
`
`
`
`
`
`
`
`
`r
`
`Twente.
`
`TIOS (
`
`mp, P.O.
`teceive ‘
`
`?
`
`B
`ae MA
`{
`
`9
`
`0)
`
`AL
`
`i
`
`é
`
`\V
`
`Abstract
`
`
`
`
`
`
`
`
`
`
`
`
`
`This paper discusses the problem of transcoding as
`
`
`
`a satellite transmits an MPEG-compressed video s
`However, since the cable c
`only limited, t
`
`
`
`> MI
`is ho.
`r¢
`
`
`
`pecific transc
`0
`
`sucl
`p
`
`
`
`;
`
`
`
`
`
`
`
`
`'
`
`in
`
`h
`
`
`
`
`
`
`
`
`
`
`I
`
`1
`
`I
`
`t
`
`{
`
`C
`
`sitt
`ita
`g
`
`
`fo
`
`It
`
`is
`
`i
`
`
`inst
`in,
`for
`it may occur
`
`
`
`
`
`
`
`9 Mbit/s. This si
`
`
`“ad end will want
`lower
`t
`1
`
`
`
`
`
`
`
`
`
`sed vide
`1
`a mpressed
`
`
`
`
`
`m d
`is
`|
`at
`
`show
`
`cal
`+t
`
`|
`
`:
`
`S
`
`1
`
`
`
`trans!
`
`
`5
`
`
`
`j
`
`
`g
`
`|
`
`
`s
`
`Tt
`estig
`
`
`
`
`
`‘
`
`3
`
`f
`
`
`
`
`
`v
`
`
`
`5
`
`
`
`SES
`
`II
`
`
`
`
`
`
`
`
`
`
`
`
`I
`
`é
`
`
`
`
`
`‘
`
`
`
`
`
`
`
`
`
`
`§
`
`NSS
`
`s
`
`t
`
`
`
`the
`
`
`
`
`
`
`\
`
`
`5
`
`
`ext
`
`k
`
`i
`
`
`sses
`
`$5
`
`
`
`
`
`
`
`
`rad
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`Vide
`
`
`
`
`)
`
`
`wit
`
`
`
`
`
`
`
`
`
`
`
`1
`
`
`
`
`
`}
`
`
`n; MPEG:
`
`
`
`
`
`
`
`T
`
`
`
`
`
`
`
`
`
`
`
`
`
`S
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`itior
`
`
`
`
`SNR wi
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`Bg
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`of bit-rate conversion is
`
`ssed
`he incoming con
`
`ed intoaco
`peration ol
`
`
`
`
`
`
`
`
`is called
`
`
`
`
`
`
`
`
`
` transcoding and a ¢
`his conver-
`
`
`
`
`
`
`
`
`
`
`
`sion 1s called a transcoder.
`In this paper the specific
`
`
`
`
`
`
`
`
`transcoding
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`(2P0D “Sf ZT ala) Mme] IyBIUAdOD Aq pa}da3}0/d aq Aew jelsayew SIUL
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`so b
`it-rat
`
`it-ra
`1LiL
`
`
`
`
`
`
`netu
`i compressed rmat
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`c
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`ite
`
`lower
`
`a
`
`7
`
`
`
`
`
`
`Page 4 of 23
`
`

`

`T = Transcoder
`
`Videoin [~ Ry —7R
`|] Video out
`9 E ft} T Lee os
`— Ld}
`}
`
`{
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`Fig.
`
`
`
`
`
`
`
`|. Basic configuration of a system including a transcoder.
`
`
`
`
`
`
`
`
`
`
`
`signals.
`Transcoding by means of cascading a decoder
`and an encoder may lead to several types of prob-
`lems. In this paper two problemsare described. The
`first problemis that of complexity. It will be shown,
`in a way similar to that of Morrison [4], that the
`
`
`
`
`
`
`complexity can be reduced significantly. The trans-
`coder discussed in this paper is, however, slightly
`
`
`
`
`
`
`less complex than that of Morrison. Anotherdiffer-
`
`ence with respect
`to Morrison et al.
`is that the
`
`
`
`
`
`
`transcoder discussed in this paperwill be applied to
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`MPEG-compressed signals instead of H.261-com-
`
`pressed signals as in the case of Morrison.
`
`
`
`
`
`
`
`
`
`The second problem is that of performance, that
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`is, picture quality at a given bit-rate. The introduc-
`discussed. Since MPEG video compression(see e.g.
`
`
`
`
`
`
`
`tion of a transcoderin the transmission chain may
`
`
`(2. 3, 5, 6, 8. 10, 12]) is the current mainstream in
`
`
`
`
`
`
`
`introduce extra losses as compared to compressing
`
`compression technology this paper will concentrate
`
`
`
`
`
`
`
`
`
`
`directly, called direct coding. to the lowestbit-rate.
`
`on the transcoding of MPEG signals into MPEG
`This can be investigated via two approaches:(1)
`
`
`
`
`
`
`
`
`
`
`signals.
`theoretical analysis and (2) computer simulation.
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`Basically, the situation with transcoding is as
`
`
`
`
`
`
`
`
`Signals passing through transcoders are typically
`
`shownin Fig. 1: an encoder compresses the incom-
`
`
`
`
`
`
`
`
`quantized twice. This cascaded quantization is —
`
`ing video signal at a bit-rate of R, (Mbit/s), then
`
`
`
`
`
`
`
`a source of performance loss in the case of trans-
`
`
`
`
`
`
`
`
`
`
`
`this compressed signal
`is converted into a com-
`coded signals. Besides theoretically analyzing cas-
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`pressed format of a lower bit-rate R; (Mbit/s) and
`
`
`
`
`
`
`
`
`
`caded quantization it
`is also possible to simulate
`
`
`
`
`
`
`finally a decoder decompresses the incomingsignal
`a complete transmission path including a trans-
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`and displays the resulting video signal. Transcod-
`
`
`
`
`
`
`
`coder. The latter does not involve any model ab-
`
`
`
`ing May occur in situations where one means of
`straction, but mayyield less insight into the mecha-
`
`
`
`
`
`
`
`
`
`
`
`
`
`signal transport interfaces another meansofsignal
`nisms which actually cause performanceloss.
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`This paperis organized as follows. Section 2 will
`
`
`
`
`
`
`transport.
`
`
`
`
`
`
`
`
`
`Essentially, a transcoder consists of a cascaded
`discuss the problem of reducing the complexity of
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`decoder and encoder, as shown in the lowerpart of
`the transcoder. The problem of picture quality loss
`
`
`
`
`
`Fig. 1. In the situation described in this paper the
`
`
`
`
`
`
`
`will be discussed in the following three sections.
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`encoder embedded in the transcoder is only com-
`A typical source of performance loss in a system
`
`
`
`
`
`
`
`
`
`
`
`
`
`pressing at a lower bit-rate than the incomingbit-
`including transcoding is the double occurrence of
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`rate, so there is no other reformatting involved,
`a quantization operation. Thefirst step is to ana-
`
`
`
`
`
`
`
`
`
`such as resampling.
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`lyze quantization—straightforwardly,cascaded
`
`
`
`In the literature the problem of transcoding
`which will be discussed in Section 3. This section
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`seems fairly uncovered.
`In [9]
`the problem of
`will present an analysis for single coefficients so
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`PCM-ADPCM-PCM lossless transcoding is dis-
`that insight into the basic mechanisms can be ob- —
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`cussed, Morrison et al. [7] describe a transcoder
`
`
`
`
`
`
`
`
`
`
`tained and this section will present an analysis for
`
`
`
`
`
`
`
`
`that has a limited complexity for video signals
`
`
`
`
`
`
`
`
`
`the combination of 64 coefficients. Section 3.4 will
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`compressed with hybrid DCT systems. Morrison et
`discuss the second step, which is to enhance the ~
`quantization model with prediction loops. Results—
`
`
`
`
`
`
`
`al. compare the picture quality in a situation when
`
`
`
`
`
`
`
`
`
`
`
`
`transcoding from bit-rate R,
`to R, with the situ-
`
`
`
`
`obtained in computer simulations of a system in-—
`
`
`
`
`
`
`
`ation where the video signalis directly compressed
`cluding transcoding will be discussed in Section 4.
`
`
`
`
`
`
`
`
`Finally, the paper will be concluded in Section 5. _
`
`
`
`
`
`
`
`at R;. Their measurements indicate an extra loss
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`i’
`
`
`
`—
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`Page 5 of 23
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`Page 5 of 23
`
`

`

`ture is denoted by x then the displacement is here
`In the introduction a transcoder was pictured as
`denoted V(x). The motion-compensated signal
`is
`denoted S[/{"),,W] in Fig.2 and the operation
`a cascaded decoder and encoder. Straightforward
`cascading of the decoder and encoder, however,
`S({-] will be referred to as the shift operation. The
`may lead to far too complex transcoders. In this
`motion compensation yields the prediction P, =
`SCI). B] as follows:
`section it will be shown that the transcoder can be
`
`
`
`
`
`
`
`
`
`
`
`
`much simpler, i.e. the number of (I)JDCTs and the
`
`P(x)=SUM, KI] @)
`amount of memory required can be reduced by
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`Il
`combining the decoder and encoder in a more
`EY be + VG).
`
`
`
`
`
`
`efficient way. The approach will be to cascade a de-
`
`
`
`
`
`
`
`
`
`
`
`
`
`coder and an encoder and then analyze which ele-
`According to the MPEG syntax, the displace-
`
`
`
`
`
`
`
`
`
`
`
`
`
`ments of the two can be combined.
`ments F(x) for each pixel are equal for pixels within
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`The transmission chain being studied comprises
`a macroblock. The macro-block vectors are trans-
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`two cascaded encoding and decoding operations.
`mitted in the bit-stream. So, besides quantized
`
`
`
`
`
`
`
`
`
`
`
`
`
`The variables associated with thefirst encoding and
`DCTcoefficients, motion vectors are also decoded
`
`
`
`
`
`
`
`
`decoding operation will have superscript (1) and
`from the incoming bit-stream. Furthermore,
`the
`
`
`
`
`
`
`
`
`
`
`
`the variable associated with the second encoding
`so-called macro-block modes are also decoded
`
`
`
`
`
`
`
`
`
`
`and decoding operation will have superscript (2). In
`
`
`from the incoming bit-streams. These macro-block
`
`
`
`
`
`
`
`
`
`system blocks subscripts | and 2 will be used to
`modes contain information about, for instance, the
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`denote this difference.
`intra/inter decision or the forward/backward/inter-
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`is
`Fig.2 shows the outline of a decoder that
`polated decision [3, 10].
`
`
`
`
`
`
`
`
`
`
`
`present
`in the transcoder. First
`the bit-stream is
`A significant
`reduction in complexity can be
`
`
`
`
`
`
`
`
`
`
`
`decoded (VLD), yielding the value of the quantized
`achieved if the cascaded encoder uses the same
`
`
`
`
`
`
`
`
`
`
`
`
`
`coefficients, next
`these quantized coefficients are
`picture types as the incoming signal. So, when a de-
`
`
`
`
`
`
`
`
`
`
`
`de-quantized (DQ,) and put through an inverse
`coded [ picture is again coded as an I picture,
`
`
`
`
`
`
`
`
`
`
`
`discrete cosine transform (IDCT). This process
`a decoded P picture is again coded as a P picture
`
`
`
`
`
`
`
`
`
`
`
`
`yields the residual signal R‘’ of the picture /‘!’. The
`and a decoded B picture is again coded as a B pic-
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`picture can be reconstructed by adding the predic-
`ture, the transcoder complexity can be reduced due
`
`
`
`
`
`
`
`
`
`
`to two factors.
`tion which results from the previously decoded
`
`
`
`
`
`
`
`
`
`
`
`
`picture J)’, by applying motion compensation
`First ofall, the decoded pictures are notyet in the
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`(shift).
`proper order. That is, although B pictures are de-
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`Let K, denote the vectorfield of picture n. In fact,
`codedafter the corresponding anchor frames, they
`
`
`
`
`
`
`
`
`
`
`
`
`
`each component of the vector represents a displace-
`are displayed between the two anchor frames. This
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`6)
`n
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`Ail
`n
`
`| V
`
`
`
`
`VLD }—» DQ,
`
`
`
`|
`
`=
`
`bits
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`ae
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`[ IDG te
`+
`
`
`
`
`
`
`|
`be a
`
`
`
`
`
`
`
`
`
`
`
`(orp ee
`Pa
`~ S|ln_ al
`
`
`
`Shift Ke MEM
`Ny \ (1)
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`video
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`Iya
`
`
`
`
`
`
`Fig. 2. Outline of the decoder in the transcoder.
`
`
`
`
`
`
`
`Page 6 of 23
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`Page 6 of 23
`
`

`

`+
`’
`————_—— I
`video
`:
`
`
`
`Prediction ©
`(2) _
`.f
`(2)
`Po”
`s 8]tn 1p] |
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`hy (motion vectors)
`Fig. 3. Outline of the encoder in the transcoder.
`
`so-called picture reordering is performed inversely
`in the decoder. The picture reordering in the de-
`coder andthepicture ordering in the encoder can-
`cel out when the two are cascaded. Here a large
`amount of memory can be saved.
`
`
`Secondly, most of the decoded information. such
`as motion vectors and macro-block modes, can be
`
`
`
`
`reused in the cascaded encoder. Especially, the ob-
`viated need to estimate motion vectors again red-
`uces the computational complexity of the trans-
`
`
`
`
`
`codersignificantly.
`
`
`
`
`
`
`
`
`At this point there is no need to have decoded
`
`
`
`
`
`
`
`
`pictures available in the transcoder because most of
`
`
`
`
`
`
`
`
`
`the incoming data can be copied from the decoder
`
`
`
`
`
`
`to the encoder. In the following it will be shown
`
`
`
`
`
`
`that, because there is no need for decodedpictures,
`
`
`
`
`
`
`
`the amount of memory forstoring the previously
`
`
`
`
`
`
`
`decoded pictures can also be reduced.
`
`
`
`
`
`
`
`The typical MPEG-2 picture level features (see
`
`
`
`
`
`
`e.g. [3]) can also be copied from the incoming
`
`
`bit-stream to the encoder. This holds for features
`
`
`
`
`
`
`
`
`
`such as
`‘picture—structure’,
`‘top_field_first’,
`‘re-
`
`
`
`
`
`
`
`peat_first_field’ and ‘progressive_frame’.
`
`
`
`
`
`
`
`
`Fig. 3 shows the basic outline of the encoder.
`
`
`
`
`
`
`
`
`
`First
`the prediction P!?’ is subtracted from the
`
`
`
`
`
`
`
`
`incoming pictures /\!’, which yields the residual
`
`
`
`
`
`
`
`signal Rj’. This residualsignalis put through a dis-
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`7
`
`—
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`crete cosine transform (DCT) and a quantizer(Q,).
`
`The quantized DCT coefficients are compressed
`
`and then outputted. Also, the quantized DCT coef-
`
`
`ficient are again dequantized and are further fed
`through an IDCTto yield the reconstructed resid-
`
`ual picture R®@) +E?) where E'?) denotes the
`
`quantization error in the second encoder, ie. the
`
`
`
`
`encoder in the transcoder.
`In the next step the picture /'?’ is reconstructed
`by adding the same prediction as was subtracted
`
`
`
`
`
`
`earlier in the process. This prediction is obtained by
`
`
`
`
`
`
`applying the shift operation S{-] to the previously
`
`
`
`
`
`
`
`decoded picture.
`
`
`
`
`
`
`
`
`
`The basic transcoderconsisting of a decoder(see
`
`
`
`
`
`
`
`
`
`Fig. 2) and a cascaded encoder(see Fig. 3) is shown —
`
`
`
`
`
`
`
`
`in Fig. 4. Observe that only the prediction loop of
`
`
`
`
`
`
`
`
`the decoderis shown. The transcoderof Fig. 4 can
`
`
`
`
`
`be simplified by removing one of the two picture
`
`
`
`
`
`
`
`
`stores (MEM).
`
`
`
`
`
`
`
`This can be done using the fact that the shift
`
`
`
`
`
`
`
`
`operatoris linear, that is,
`
`
`
`
`
`
`
`
`
`
`SUM, + £24) <
`
`
`
`
`
`
`
`
`
`()
`SEIS? Yi) + STEP), Ka
`
`
`
`
`
`
`
`
`
`
`
`
`holds. Usingthis linearity property the prediction —
`
`
`
`
`
`
`
`
`
`in the second encoder, i.e. the encoder within the
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`Page 7 of 23
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`Page 7 of 23
`
`

`

`G. Keesmanetal.
`
`
`
`
`
`Signal Processing: Image Communication 8 (1996) 481 S00
`
`
`
`
`
`
`
`
`
`
`485
`
`
`
`Paint X
`
`
`
`Point Y
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
` ae
`
`
` sf) v,Jest.y,) s[ey”gp (2) Yq]
`ae}Yn J
`
`
`
`(2)
`(1)
`= In +E,
`
`(2)
`hh
`
`
`
`
`
`
`
`
`Fig. 4. Cascaded decoder and encoder as transcoder.
`
`
`
`
`
`
`
`
`
`
`
`transcoder, can be split into two components. One
`
`
`
`
`
`
`
`
`of the two components in the encoder prediction
`
`
`
`
`
`
`
`results from the motion compensation applied to
`the previously decoded picture /) ,. Observe that
`
`
`
`
`
`
`
`
`this prediction is first added in the decoder at ‘point
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`X’in the figure and is then subtracted in the second
`
`
`
`
`
`
`
`
`
`encoderat ‘point Y’ in the figure. These two opera-
`
`
`
`
`
`
`
`
`
`tions of course cancel one another.It is only neces-
`
`
`
`
`
`sary to subtract the motion-compensated quantiz-
`
`
`
`
`
`
`
`
`
`ation error, as will be shownin the following.
`Using Eq. (1),
`the residual R‘ in the second
`
`
`
`
`
`
`
`
`encoder can be written as
`
`
`
`
`
`
`
`
`
`
`
`RO = 1 — SUM + ES]
`
`
`
`
`
`
`= 1) — SU, WI) — SLE, Ma.
`
`
`
`
`
`
`
`The decoded picture I!’ can be written as
`
`
`
`
`
`
`
`
`
`
`
`I) = Ry + STL, Ki.
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`Substitution of Eq. (3) into Eq.(2) yields
`
`
`
`
`R® = RY — SCE? i].
`
`
`
`
`
`
`
`(2)
`
`
`
`(3)
`
`(4)
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`All that is needed to be able to produce the out-
`going bit-stream is the residual R'?). According to
`
`
`
`
`
`
`
`Eq. (4), this residual R!?’ can be obtained directly
`
`
`
`
`
`
`
`
`
`
`
`
`by subtracting the motion-compensated quantiz-
`ation error E’?),. Hence, only E’?) need be stored,
`
`
`
`
`
`
`
`
`
`
`and the store used for J’ can be dropped.
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`Fig. 5 shows the block diagram resulting after
`
`
`
`
`
`
`
`
`
`the removal ofa picture store. To obtain the coding
`error E’?) the input and the output of the coding
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`path consisting of DCT, quantizer, de-quantizer
`
`
`
`
`
`
`
`and IDCTare subtracted. Consecutively the coding
`error Eis stored in a memory. The residualsignal
`
`
`
`
`
`
`
`
`
`Ris obtained by subtracting the motion-compen-
`
`
`
`
`
`
`sated error S[E,, ¥,] from the decoded residual
`
`
`
`
`
`
`
`
`
`
`RW.
`
`
`
`
`
`
`
`Another simplification can be realized in the
`transcoder of Fig.5 by moving the DCTs and
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`IDCTs around in the block diagram. The DCTs
`
`
`
`
`
`
`
`
`
`and [DCTs can be moved around by using the
`
`
`
`
`
`
`
`
`
`
`linearity property of the DCT. Fig. 6 shows some of
`the possibilities. The top section of the figure shows
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`that in a data fork the DCT (or IDCT) can be
`
`
`
`
`
`
`
`
`
`moved from the main path to the two separating
`
`Page 8 of 23
`
`Page 8 of 23
`
`

`

`from VLD
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`|
`)+ OF
`—— ——s
`to VLC
`+}00) IDCT ree +7} BCT ae Q20L- ——p
`
`
`ee
`“feShift
`
`sfe vy].
`
`
`
`
`
`
`
`Fig. §. Transcoder with a reduced amount of memories.
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`pe,
`
`(IJDCT —e
`
`
`
`(DCT
`
`i
`+ ()DCT |——»~
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`—g! wocr |
`fera|q
`Soe
`—[eae|
`
`aLe ea
`POUpet|=
`
`
`
`
`i|
`
`DCT
`
`
`
`|
`
`Fig. 6. Moving DCT and IDCT operations around.
`
`paths; the middle section showsthat at an addition
`
`
`or a subtraction the DCT (or [DCT)of the compo-
`nents of the addition can be moved to the DCT (or
`IDCT)of the result. Finally, applying an IDCT to
`a signal that has passed through a DCT equals
`a unity transfer. This property is used in the lower
`
`
`
`
`
`
`
`
`
`
`part of Fig. 6 in combination with the property for
`
`an addition.
`
`Moving the DCTs and IDCTsof Fig. 5 around
`in this wayyields the transcoder of Fig. 7. Observe
`that the number of DCTs and IDCTshas beat
`reduced from three to two. At this point the com=-
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`Page 9 of 23
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`Page 9 of 23
`
`

`

`from VLD to VLG
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`(-pictures
`
`A
`i
`
`
`
`Fig. 7. Transcoding system with a reduced number of DCTs.
`
`
`
`
`
`
`
`
`
`
`
`
`
`Shift LH Mem ~—
`
`B-pictures
`
`~O
`
`
`
`
`
`
`
`
`Previous
`
`
`
`
`
`
`
`
`
`~
`
`
`
`The whole procedure followed for deriving the
`
`
`transcoderofFig. 7 is implicitly based on the use of
`i
`P pictures. The derivation is, however, equally valid
`|}
`Previous
`for I and B pictures, which will be shown next.
`
`
`
`
`
`
`|
`Shift — Mem
`Fig. 8 shows the decoder prediction loop for I,
`zero
`Next
`B and P pictures. When anIpicture is decoded, the
`
`
`
`