Second Workshop on
`
`cp :!i! ~ ~ iH'!I
`
`cp 5I: fiJI ~ ~Ji:
`'I'HJ1. I ~ ~ ll ~ I ~ t.Uii fiJI ~ ~Ji:
`~~
`J=~m I!OJM fil fllJ 1:<!"' illiMfil · {§ <l';l!oH ~~!I< <P ;;o 'l!li'l il 'Til'! J!l!!Uii 19 96 'I' J:ll =Ill TIll
`
`

`

`l'!p at~ -9! 1/t a,. Ht ~ ~t 1t J
`%-A it 1\ ;f<.. 1
`(Real-time and Media
`Systems Workshop' RAMS'96) .i..J!JtJf..{?L • {.;J~U~1t~1.::.*-~:l.'iJJt
`·~·~·~··k*A~~*~o~-~#lt~·~#t~±~~
`•••·m~~«•~~·~~ttma,.M~~-~~~~•&·•~
`a,.MT~·a•~*&#·~~-~·t·T···~····o~
`~~tlt-a!J m ~~k*•-;k. • • t~M:. · ..:r..~~ · t-*1t1t ( 85•7}]
`1 a f), -m k <t1t~RJ) · Jt Jltlt-~ ~ .i..¥lt · *-lt- • t~~-fn-ft1~~
`ma•t~-1@~5t,t.~ "fi-i-¥lf; ~-•klt-~~•7fl 30 a· 31 a~
`M·k•t~~·..:r..~~~••Jnt·+•m•t~~-fnt-*•~~•
`~jlfo
`
`~~·~·*·~~#~·&·~~A&~jf..{?L~~·-hT•
`~~+~~·~~·o···-~~--~*···~~~!-T•
`'f f),~ 11 ~ ;f ..t J!JJ f~. :f<.. -tiil X. A .i.. ~ El •
`•. ~a,. M t 11' ~ ti11f! ft Alfi Jl:.
`~•:t·T•••~~~•*~~~•x.•*-•a•••~~#~
`Mt·~--~~~~TmAm•~a,.·~~·~••·~~e~•
`&;~r+~·••~*lt--~~J~;t..tJ!Jlfl**kt~*•~•
`X.·~~-h~-a,.MT~~-T~~~··~t·T··-~~1
`~m•~~-~~~·~~•••~-~~~~Mo~~aaa,.MT
`~~~•~·t·T•~••§+~Jtm-m~~kh3•~•~~{.;J
`~-·~~~Jk.·~~~#lf~.i..~EJ~o·~~klf···&~-­
`~A~*•*••A~~~~-EJ-~~T+~£~~-·0
`
`A~~~~~--~100.i..¥ltJf-1ft~kh~M·t-ftkt~~Jt~
`~&t-*t~~~ffl~~~~~~•aAT~~~~·a•&~~·*
`~~~k•~~~J!JJ•~~MoA~~~••*e~~~--~••
`
`~ jj 11'5 ~ + 1} . ' * tit #fl:l:dd!:. il* * ~ ~ '£" ~ • :f.¥. 0 :l-1! ffi ~X. k lt IBJ ~
`
`~~·~ffi~~--lf~~·J!JJ-*o
`
`J
`
`

`

`::k"fi"!l! • .±..JNi' (Honorary Chair)
`*:!!14-
`::k 1t ~ J61.±..JN; ( General Co-Chairs )
`*J#.J.J>..:£.1:-±.
`
`::k1t~~.±..JN; (Vice Chair)
`.:JlAf. "i"
`
`tf:1jl~ Jit 1t(Steering Committee)
`
`1:-~~ #~~ $~~ ~~# ~~~
`.ftM. ~·t *~~ aT~
`~--
`
`..:E.. g.. **• #-· ~~~
`••M ~~M -~~ -~#
`
`~~!.~A 1t(Program Committee)'
`
`-
`
`i
`
`"Pl:ff~.il'.'JHil"lr~.-:E
`
`111111111111 1111111111 IIIII lllllllll/llll/11111 111111111111111 111111111111111111
`3 0330 00 011406 1
`
`~J6J.:t.J$'
`f.irJt~(<f:Yc'"JfttJltwWt.M-l
`%1 .l ~ ( xi!~ -t1t'"Jfttrlf l
`*~~ ( i:~*-*wWt.~ l
`~.i. _. A
`•
`~(<f:Yc'"JfttJltwWt.M-l
`..:EJt~(wWt.~~~~~l
`..I.
`~-~(•**-*~Wt.~l
`*~~(xl!~t~'"JfttM-l
`**~(xi!*-*w~~l
`~~~(~'"Jf~ti!M-l
`#~~(•~••)
`*-~(<f:Yc'"Jf~JltwWt.M-l
`~*~(wWt.~~*~~l
`~~~(<f:Yc'"Jf~~wWt.M-l
`::t';tii ( xl!*-*w#~ l
`~~Jll ( xi!*-*w#~ l
`•*-•(<~'~*-*w#~l
`•~£(xl!~t~'"JfttM-l
`~*~(~~*-*wWt.M-l
`~~~(<f:Yc'"Jf~~wWt.ml
`~~~(<f:Yc'"JfttJltwWt.M-l
`~*•(<~'~*-•••~)
`~~~(<fJ.J#.Jlt)
`~-~(•**-*·#~)
`~~ffi~(i:~*-*wWt.~l
`~T~(~'"JfJltti!m-l
`iat*-"t ( :k.fiil~·Jlt·Wl.~)
`•~~(i:~*-•••~l
`
`·i¥1~ ( ~J'}J;k_···~)
`
`•~~(xl!~t~'"JfttM-l
`
`111
`
`

`

`~U: .. til ( 'f JJ :k.. * 1t _r.. ~ )
`.~ ;it R. ( _r.. -Bt Jl7t 't :i@ .PIT )
`
`.r. i"F 4- 1\ 1t(Operations Committee)
`
`:;:;:f':11~ -1T"
`
`'* ~;!g ~
`P-*..i£~ ' ~~)If ' )$.~til ' Ji ~ '~
`.J..:fk.~ ' .1.. :k..J.b ' ~~if
`.1.. lL'~a}]it,~-k#
`
`iv
`
`

`

`7J'l3oa
`
`( !l J!JJ_::_ )
`
`9:00 Opening
`(ROOM A)
`
`9:20
`9:20 Invited Talk (I) ~*-•t
`r 'tH$11ll13-i¥:1lll~iHI'99!il~J
`.i.#A..:.l.1t.±.
`(ROOM A)
`
`10:20
`
`10:20
`
`Coffee Break
`
`10:40
`10:40 SessionAl
`Session Bl
`Session Cl
`51% r ~p 1M' J!< .1:. it 51-. '~il!.ill."lltJ: 'f ;t 'A'"lltlt>£#.
`!;OJJ3-qtllf J : A2, B2, B3, B4, B5
`{If
`.i.#A..:ffii£;1) .i.#A..:.l.Jt~
`A3, A4, A5
`.i.#A.. : ~.t, :t
`11:05
`11:05 Session A2
`~ 1M' ill it Rllr
`.i.# A. : -t ~ t<>
`(ROOM A)
`
`Session B2
`ll!il!. ill, lii.(I)
`.i.# A.: §.:PI; Jf JE
`(ROOM B)
`
`(ROOM C)
`
`12:20
`12:20
`
`Lunch
`
`13:20
`13:20 Invited Talk(II) Bao Hwang
`
`r Internet. Intranet and Infonnation Infrastructure J
`.i.#A..: i'[,t,.Jt
`(ROOM A)
`
`14:20
`14:20
`
`14:40 Coffee Break
`14:40 Session A3
`Session 83
`Session C2
`~p 1M' A\ M. £ >!! q§.
`;l;.i!i.ftittl.J;ff
`Jl!ilill.tlt(II)
`.i.#A..:I't,~-f .i.#A..:-t-t.<;> liT
`(ROOM A)
`.i.# A. : l't.J%1oli
`(ROOM B)
`(ROOM C)
`
`16:30
`16:30
`16:50
`16:50 Panel (I)
`
`Coffee Break
`
`HUt 1tffl IM'1\~1 t-IIUo fr!] lJ!J ~
`
`.i.#A.:"'1!!:~
`
`(ROOM A)
`
`Reception Cocktail
`
`18:50
`18:50
`
`20:30
`
`7 J'l 31 a ( !IJ!Jl .=:.).
`Session B4
`9:00 Session A4
`Session C3
`51 a r ~lit 'It' lift
`ill, t1t -t-tk A\ J.U .. ~it ill, t1t (Ill)
`www
`liT J : C4, C5
`.i.#A..:*f.iS-f\
`.i.#A..: ~{;:..~
`9:15
`'9:15 (ROOM A)
`Session C4
`~lit 'If JiqUf(l)
`.i.# A. : jjl1~ Pili
`(ROOM C)
`
`.i.#A..:-t#*f.
`(ROOM B)
`
`10:45
`10:45
`
`Coffee Break
`
`11:05
`Session B5
`11:05 Session A5
`q§. {If ill. tt
`q§. {If ili.li
`c II! it ill. -m.J
`r ~p 1M' N .1:..i!llll
`.i.#A..:_ffii£;1)
`13-Jtlir J
`.i.#A..: ~.t,:t (ROOM B)
`(ROOM A)
`
`Session C5
`Jij;ffJll.tJ[
`r itJ!t 'lt'~Ji {If J
`.i.#A..: ~{;:..:§
`(ROOM C)
`
`12:25
`12:25
`
`13:30
`13:30
`
`Lunch .
`
`Invited Talk (Ill) Albert Teng
`r Intel's Vision of Multimedia and Communication
`Applications J
`.i.#A..:~~ ..
`
`( ROOMA)
`
`14:30
`14:30
`
`Coffee Break
`
`15:00
`15:00 Panel (II)
`
`fo t¥:1lll1,ffl!M'f>;A>.m 9911lln 'It ,JCJ!l-!illl
`
`.i.#A..:.l.i;.±.
`
`I7:00
`(ROOM A)
`17:00 Business Meeting
`18:00
`(ROOM A)
`18:00
`
`Banquet
`
`20:30
`
`v
`
`

`

`Invited Talk (I)
`'it&\l~~~U~-!-f!IJJl~
`~~~ < <~> m at:¥a.) ..................................................................................................................... 1
`
`Invited Talk (II)
`Internet, Intranet and Information Infrastructure
`Bao Hwang (Naval Surface Warfare Center, U.S.A.) ................................................................... 3
`
`Invited Talk (III)
`Intel's Vision of Multimedia and Communication Applications
`Albert Teng
`( Intel Corporation, U.S.A.) .................................................................................... 5
`
`Panel (I)
`~~·-ft.~at~;f:l:.-t-m.~f!IJ~~
`.i.#A:~iltbl!.
`'JI ~A: 14-~ ( .:;LiHt) , ~'!"~ ( 'i>:!k:~JtBE;f:I:.#Pif) '
`~~~(<f:!k::~%81;) ,g*~<<F*it~~~)'
`* ~<~~~~*~"*") ,~~~<~~x**~Pif),
`
`~~~(~~~~~~-t') '~~~(.:;!:.*~)'
`~~~<~~x*~Mml ,~~~<~~~~*~"*"),
`~ ~(~~~) ,·~~(<f:!k:~JtBE;f:l:.-t'Pif),
`•••<<F:!k:~JtBE~~Pif) ,~-~(<f:!k:~JtBt~APif) ............................. 7
`
`Panel (II)
`~~-~~*~~~f!IJ~~~*~ll~
`.i.#A:I~±.
`iJI ~A : *-It~ (:II.!-) , ~.iE~ (.f. ifni) ' ;f1..£-M ( Seed net ) '
`~~~(·~~JtPif) '~~(it~~~)'
`~~~(<f~at:¥il.) '-~~(~~)
`jlif f.J 4'- ( :fu1t ~;fl. iljl )
`, Bao Hwang ( Naval Surface Warfare Center, U.S.A.)
`~~~<«~•~) ,~•~<<F.iEx*l
`*w•<**x*) ..................................................................................................... 9
`
`vii
`
`

`

`51~ : l!p llt~~:i!~l!ift#r
`SessionAl
`.i..#A: .ft.~%
`
`July 30, Tuesday
`
`s
`
`Session A2
`
`l!p at i! 1-t ft. #'f
`
`.i..#A: *' ~t,:.
`
`ISDN; lt:i!~x#UUic-i:..llp Jlt;W; ;jt~~it
`J1 El ;!$;' ~*~ '*1t"'' '-j,j~H}]· ( t-*'t1"t~a)) ................................................... 11
`Real-time Distributed Clock Synchronization over Ethernet
`Mei-Wen Li, Chin-Fu Ku, Yu-Chun Pan, Chia-Hui Wang,
`Meng-Chang Chen, Jan-Ming Ho, Ming-Tat Ko (Academia Sinica) ..................................... 19
`On Supporting Real-time Channels over Ethernet
`Kuo-Hui Tsai, Chia-Hsiang Chang, Mei-Wen Li (Academia Sinica) ..................................... 27
`Real-time PC-based Software Implementation ofH.261 Video Codec
`Den-yuan Hsiau, Jiun-Jie Huang, Jiun Shiu, Ja-Ling Wu (National Taiwan University) ......... 33 S
`A Novel Medium Access Control Protocol and Its Implementation for Wireless PCNs
`Hsiou-Ping Tsai, Yen-Yuan Chiang, Chung-Ping Chung, Kong Dar Fan,
`Chia-Chi Huang (National Chiao Tung University) .............................................................. .41
`
`51 ~ : M_i!;#l. iR.
`Session Bl
`.i..#A:~~~
`
`Session B2 M.i!& ~(I)
`.i..#A:~&.~ffi.iE.
`Real-time VOD System Based on Multi-Port Network and Hierarchical Storage Architecture
`Chui-Chu Cheng, Jui-Chi Chen, Chi-Yui Chiou
`(Industrial Technology Research Institute) ........................................................................... 49
`Design and Implementation of a Multimedia Storage Server Using
`Parallel Accessing Disk Arrays
`Eau-Chung Lee and Tien-Fu Chen (National Chung Cheng University) ................................ 59
`a#EN:am~&~••~•·~~~~m~•••~•#••:i!~W#~
`W#:&ii~i\.
`~·~--~·-*~~-~··(4~*·-~·ft·~~~) ............................ ~
`~-tii\.; ~ft'A'#-m ~ ~-'lr-1~1it!l'.
`
`-~--~~~(••*•) .......................................................................................... 75
`
`; ~d:10JJlll2--tH.. fie. jj ~1H"-
`-Jf1.:: . .;;t' ~*Hi ( 'fiE.*•) .......................................................................................... 85
`
`viii
`
`

`

`tf X it-mtAt~;it.fij-
`Session Cl
`.i..#A..: .. ut~
`Document Analysis and Recognition System
`Fu Chang, Trai-Fu Chiu, Tzren-Ru Chou, Min-Ching Lee, Ya-Ching Lu,
`Tsuey-Yuh Shuai, Tzu-Ming Tan, Jian-Jon Wu, and Chu-Sing Young
`(Academia Sinica) ................................................................................................................ 91
`
`i4fi~ + ~ 7'6 <lf! * liAJt~ttAt~;!J€.}1.
`
`i'Jlt#.tt, 1Mt.lil (it-**-*) .......................................................................................... 99
`t x*-tH: .. *;t•tU~1Jf~~ik-&..Ji:.J.IUJ
`iJ!R::;t;i;lt, ~1t.ft, 1Jlill:if.,;, ~.r.Jtilf:, .:I:i*~ , 1Ull3.t ( ..:r. t:?t#r-OJfJl:M;) ........... 107.
`4i!'it!3t!tx*i\.~.J.ll!-mt2--t-"Jf%
`•~•,•&•,s*•,•••,~~•,•~*'*•x
`( t:J~tx<lf!, <f-*it1t.z}iiJ) ..................................................................................... : .. m
`
`~it I! 1t ;it #j(cid:173)
`Session A3
`.i..#A..:I'Jjt~-f
`t.!W!~l!J&*~~~1-ji~tl;.Jfl
`
`' 'lt ~""'' ~~J.J 'Jijl*• 'i'f:fft~ ' ~~~'
`
`..£~~' g * :t ' !R::•t€;*
`!Roil~' j'aJl €\ 'ij~.ljl:' :!!lt.li.lii'f' ·A;~' !fil'!-!1. ( <f-*it1t.z}ii)) .................. 117
`Deployment of ATM Broadband Network Trials and Services
`Lung-Sing Liang, Cheng-Sheng Lin, Yen-Ting Chen, and Rong-Ruey Lee
`(Telecommunication Laboratories) .................................................................................... 123
`. ATM $ ..:r.2-.t~tt -&.f.t-Jil
`i'f·t&OO, a.taJl, ~t.:.:f1lt, edt%', ;f1.'f:-tff, '*77M., 1~wd::.,
`·A;~' Wil'!-!1. ( <f-*i!i:1t.z}ii)) ................................................................................ 133
`Design and Implementation of a High-Speed Medical Image Transport Protocol
`over ATM Networks
`Hsiao-Lung Wu, Zsehong Tsai (National Taiwan University) ............................................. 143
`Multipoint Video Conferencing System over ATM Trial Network in DGT
`Lung-Sing Liang, Cheng-Sheng Lin, Chun-Hsiung Wang, Chung-Chi Chang,
`Rong-Ruey Lee, and Chih-Hen Lin
`(Telecommunication Laboratories) .................................................................................... 153
`ffl..f }(. &t ;t ~JI. -mt &. tlt..Z..'itf.tr" w, ~ .z.. ~tt $it~ 11
`~'*-iL. , a:~toJ.t , *1::.9t c t .iEx<lf! l ........................................................................ 161
`Routing Multicast Streams in Clos Networks
`De-Ron Liang, Chen-Liang Fang (Academia Sinica, Jin-Wen College) ............................... 169
`
`ix
`
`.11
`
`.19
`
`.27
`
`.. 33
`
`.. .41
`
`.. .49''
`
`.... 59
`
`.... 67
`
`..... 7l
`
`..... Sl
`
`

`

`Session B3 M.J!;/:11. ilt(II)
`.i.#-A.:*"*"~
`An Implementation ofMHEG Used in the DA VIC Set Top Box
`W.C. Lin and H.-M. Hang (National Chiao Tung University) .......................................... 177
`1l1l-:iJI. ill.Jlll.~-Z-!€Ult1'1' 1!- *-~
`.................................................................... 185
`F.it.7.!.(,4-t, wHJt, ~jL1. (it~ ..J:.<$i'!t:)
`x•~·~~·-~---~1-&IM-Zt~-·~
`.:JLif.'m, w<f JL, Fti\ -t, Ft1t--t c xi!*-* l ......................................................... 193
`l!f at% ·MUt 11 ltJ!l)t-Z tit. a11
`$:k..'wB·-~±--A~*-·~--~A*(<fL:k.$) .......................... W1
`Real-time PC-based Software Implementation ofLayer-ll MPEG-1 Audio Codec
`Mei-Juun Guu, J.J. Huang, Jiun Shiu, Ja-Ling Wu (National Taiwan University) ................ 209
`Performance Analysis and Evaluation of a PC-based Video-on-Demand System
`Shiao-Li Tsao, Jan-Wen Ding, Che-Ming Wang, Shiang-Chun Liou,
`Yueh-Min Huang, Gin-Kou Ma
`(National Cheng-Kung University, Industrial Technology Research Institute) ..................... 215
`
`l!f at I M.~•lf!At. .Jtt
`Session C2
`.i.#-A. : ft~JU<fl
`Abort -Oriented Concurrency Control for Real-Time Data Access
`Ming-Chung Liang, Shao-Juen Ho, Tei-Wei Kuo, LihChyun Shu
`(National Chung Cheng University, Chang Jung University) ............................................... 223
`Hybrid Online/Offline Scheduling for Hard Real-Time Systems
`LihChyun Shu, Michal Young (Chang Jung University, Purdue University) ........................ 231
`Real Time Vehicle Control and Scheduling of Multi-depot Physical Distribution System
`Chwen-Tzeng Su, Bou-Chun Kao (National Yunlin Institute of Technology) ..................... 241
`A High-Speed Real-Time Video/Graphic Controller Design
`Jing-Chang Chen, Wei-Lun Huang, Hung-Yih Hsieh, Jins~uan Eric Lee
`(Industrial Technology Research Institute) ......................................................................... 251
`Timing Expression Simplification in the VERIF AST Real-Time System Verifier
`Chia-Tien Lo, Fam Wang (Academia Sinica) ..................................................................... 259
`
`Session A·
`
`384R
`
`Telec
`
`The I
`
`Web-
`
`Session B4
`.i..#-.
`Desig
`
`A PC
`
`X
`
`

`

`7
`
`5
`
`13
`
`)1
`
`)9
`
`15
`
`:23
`
`t41
`
`251
`
`259
`
`July 31, Wednesday
`
`;#/. "#!. i-~ & #t.J.t WWW
`Session A4
`.i.#A:;Jtx;.~
`-~~~"#ti-~&#t.4.WM~M§·~~~~
`#~~--~B·#M¥··~~-A~*(~~····-~~~) .................. VI
`384K -~~ISDN :f!l."#t-lt~& .Wt.~.ft-~
`a••·a*~-•~•·•••·•~~-~~£·••A·&*ft·
`'*1Ht • *-~J>. ( <f¥t1t~~) ....................................................................... 279
`Telediagnosis System on ISDN
`Pen-Yuang Chang, Wen-Hen Luo, Sun-Lang Hsiao, Feng-Yu Hung
`(Telecommunication Laboratories) ................................................................................... 287
`The Design and Implementation of a Real-time Video Conference System
`Chia-Hsiang Chang, Wen-Chong Chang, Chien-Lin Chen, Wen-Yuan Hsiao,
`Wen-Jie Ji, Jyun-Naih Lin, Wen-Hsien Lin, Cheng-Kuo Liu, Kuo-Hui Tsai,
`Yo-Yue Wang
`(Institute for Information Industry, Academia Sinica) ............................. 294
`4~~~M&AM~&..WM~&#i.T~~"#t-lr~&#t.~t#ATa
`
`* xttt • J.l\ :it*- ( t J> :k. if: J . . ... . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . ........................................ 3 o 1
`
`Web-based Broadband Service Creation Environment
`Ying-Dar Lin, Yuh-Tay Lin, Po-Ning Chen, Michael M. Choy
`(National Chiao Tung University, Industrial Technology Research Institute) ..................... 309
`A Web-based Interactive Graphic User Interface for Netwolk Management
`Pa-Ning Chen, Tse Cheng, Jiann-Liang Chen, Ming-Hung Lin, Jiann-Hung Lin,
`Michael Choy and Bao-Shuh Paul Lin
`(Industrial Technology Research Institute) ........................................................................ 319
`
`M_~;#l. "#!.(III)
`Session B4
`.i.#A:.~;Jt
`Design of an MPEG-2 Video-on-Demand Server
`Chih-Yuan Cheng, Chang-Hsiang Liu, Chuan-Li Wang, Yen-Jen Oyang,
`Jan-Ming Ho, Meng Chang Chen and Ming-Tat Ko
`(National Taiwan University, Academia Sinica) .................................................................. 327
`A PC-Based Video Server for True VOD and Near VOD Systems
`Liang-Wei Lee, Yeou-Chyuan Chang, Yin-Hwa Huang,
`Sing I.S. Young, Huang-Cheng Huang, Wen-Deh Wang, Chien-Hsing Wu,
`Jei-Shoung Huang, and Bor-Shenn Jeng
`(Telecommunication Laboratory) ....................................................................................... 335
`
`M.~:f!l. "#!. * #t.§ fJJ ~}J ~~~~;/&_
`
`~*~·R-~£·I~<f···~--~ti·M~~--~-
`( 1;'/f:k.-*' <f:R:~~f.it) ........................................................................................ 341
`
`xi
`
`

`

`Experiences on Developing PC-Based VOD Servers
`Der-Jen Lu, Yu-Chong Wang, Jan-Ming Ho, Ming-Tat Ko and Meng Chang Chen
`
`(Academia Sinica) ........................................................................................................... ; .. 3 51. i ;:';; ;0.;~
`i:t.-{1;- i(.;'G.!Jlil. ~ ~~Jf%(HFC)T :Z..Ji..f/J i(.flltlt:?JI;Mt.mU"t *- .%
`~7'6··~4···~~·*~•·#~··~····~···~~
`( t '*it 1t -0- ~ ) "". "":""""""""""". ". """ """ ". ". "" .. """ " .. """". "" " ... ". "". "" "" ... 3 57
`:$1-{ti(. KTV *-#i.:Z.."t.£:~tcl4'1f1t
`w~.-t · tlt~tit • ~1±~ c ~tit~iHf .$.~ • t w*-* l ............................................. 365
`?l"t' : §.it '1f oltft .fir
`Session C3
`.i.#A. : ;#;.1.::..'&
`
`§lft'Jfo!tft.f.lr
`Session C4
`.i.#A. : ~1~Ji)Ji
`An Efficient Architecture and Methodology for Developing Virtual-Mall-Like
`Applications on Internet
`Pei-Sen Liu and Meng-Cheng Chen (Institute for Information Industry) ............................. 373
`
`§. •1r':l.t.ff.~ ~ ~lf% lr.J lUl 1f Jll-* VRML ii'J 'it 2-W!f%
`-·~·~~&(tW*..*) ........................................................................................ 3~
`-1JID~;fl{f_ www J:.lr.) $ A.Ji..f/J>iolt
`wt~·•~*·~•w·•~•c~~*-*l ......................................................... 3~
`
`Photo VR: A System of Rendering High Quality Images for Virtual Environments Using
`Sphere-Like Polyhedral Environment Maps
`Wen-kae Tsao, Jiunn-jia Su, Bing-yu Chen, Ming Ouhyoung
`(National Taiwan University) ............................................................................................. 3 97
`Panoramic Stereo Imaging with Complete-Focus Views for Virtual Reality
`Ho-Chao Huang, Yi-Ping Hung, and Sheng-Wen Shih (Academia Sinica) .......................... 405
`~a~•~aMA±a&~M•~z..~•Ji_•i(.••*-#i.
`~-~·#~B·*~~·wM~·•••·~·~·~«I·M*~·
`•¥*. ~1.::.~~. *-1tt ( t;Jc~%~) .................................................................... 411
`
`ft.f.lr Iii!. it: llf at~ ~.it~JI%ft.f.lr
`Session AS
`.i.#A. : ~.'t,i'
`
`ft .fir & it: &113!~JI. ill.
`Session BS
`.i.#A.:~a~
`
`ft.f.lr & it: htit1' oltft.f.lr
`Session CS
`.i.#A. : ;#;.1.::.,t,
`
`xii
`
`

`

`Photo VR: A System of Rendering High Quality Images for Virtual Environments
`Using Sphere-like Polyhedral Environment Maps
`
`Wen-kae Tsao Jiunn-jia Su Bing-yu Chen Ming Ouhyoung
`
`Communication and Multimedia Lab.
`Department of Computer Science and Information Engineering
`National Taiwan University, Taipei, Taiwan, R.O.C.
`
`Abstract
`
`In the first part of this paper, we describe a novel
`method of panoramic view rendering for real-time
`interactive applications trying to present a real-world like
`environment. This is different from the QuickTime VR
`approach, and is accomplished by generating a sphere(cid:173)
`like polyhedral environment map from photo-realistic
`images and using the generated maps to render the scene
`by teclmiques of computer graphics. In the second part,
`we purpose a simple image-based method for observing a
`certain object from different views interactively. We call
`it "object viewer." Although some precision in alignment
`is lost, the high complexity, high quality and high frame
`rate performance on a Pentium PC without any hardware
`supported makes this approach attractive in low cost
`systems.
`
`Introduction
`
`Traditionally, when we want to render a scene in the
`real world, we model the scene by geometric elements,
`such as triangles, and then render these objects. However,
`this method usually works well only when the quantity
`and complexity of objects are relatively low, often far
`below those in the real world. Furthermore, the shading
`is either too slow or uorealistic. On the other hand, the
`quantity and complexity of objects in images taken from
`a camera are almost the same as those in the real world.
`
`In recent years, some method of image-based rendering
`has been developed, such as "image warping" used by
`"QuickTime VR" of Apple Computer [I ][2], or texture
`mapping used by us [3]. These methods render the scene
`by using cylinderic or prismatic environment maps
`generated from images of the whole 360-degree view of
`the scene. However, these methods have a common
`problem: the environment map used can not cover the top
`view and the bottom view, so the vertical viewing angle
`is limited. In the following, we will propose a better
`method that renders the scene by sphere-like polyhedral
`environment maps, and thus the problem is solved.
`
`If the above problem is solved, we still have another one,
`say object viewer, for real applications. The traditional
`method for observing an object from all views is to
`generate a polygon model of the object, and then the
`model is rendered from different view points. However, it
`is difficult or impossible to generate a model for some
`objects, such as a fuzzy teddy or a valuable artifact. We
`then propose an image-based method to solve this
`problem.
`
`Texture Mapping Approach
`
`To render the scene in a room, we propose a way to put
`a camera in a proper position, such as the center of the
`room, and then take images of the whole view from the
`camera position. These images are arranged as a sphere(cid:173)
`like polyhedron consisting of textured trapezoids. This
`polyhedron forms the environment map of the scene and
`is used as the "3D model for rendering. It is clear to see
`that, when moving around the center of this sphere-like
`the user will
`the image presented to
`polyhedron,
`approximate the one rendered by traditional method used
`in computer graphics or even the one seen in the real
`world if all objects in the room are a certain distance
`away from the center.
`
`To render such a sphere-like polyhedron Is quite easy:
`no shading is required. No hidden surface problem
`occurs. (Polygonal) texture mapping is usually contained
`in rendering packages, libraries or graphic hardware
`accelerators. Thus high complexity, high quality and
`high frame rate become possible even on low cost PC
`systems.
`
`To allow a user walk around the scene, one can generate
`several such sphere-like polyhedrons by taking images
`from several different positions, each corresponding to a
`"visible area" in the scene. One polyhedron is rendered at
`a time, and only when the user is in the corresponding
`"visible area" will the polyhedron be rendered. Adjacent
`visible areas will be overlapped instead of providing a
`clear border (figure 4 ), and if a user is in the overlapped
`area, only the previously rendered polyhedron will be
`
`-397-
`
`II
`
`I
`!I
`ii
`I!
`
`

`

`rendered to avoid frequent switching of 3D models when
`the user moves around the border. We will describe this
`more clearly with an example later in this paper.
`
`The switching between 3D models will be observed by
`the user, just like the switching of acts in a film. Because
`the switching is performed when the viewpoint is moving,
`it will not be so annoying to the user. However, it is a
`problem
`remaining unsolved by us yet. Linear
`interpolation is, perhaps, not a practical solution. Not
`only because the time for rendering will be at least
`doubled (for at least 2 cylinder-like prisms need to be
`rendered), but also because the interpolation may involve
`theories in computer vision: objects in images must be
`identified and located for interpolation. The complexity
`of these interpolated images taken from the two real
`environment images usually makes this process (object
`identification) too slow and even impossible. However, if
`geometric information of the scene is available and
`objects in the
`images are identified
`in advance,
`interpolation (such as morphing) might be possible.
`
`Camera Calibration
`First. we have to measure the horizontal and vertical
`viewing angle of the camera. Then we convert horizontal
`viewing angle to the camera constant Fw, and convert
`vertical viewing angle to the camera constant Fh:
`
`~ ···.vim.ge width
`
`Fw = W cot _h_onz_· _on_tal:::...c.v':.c·em;:.=·n,_g.::an::=g:::le:___
`2
`2
`
`image height
`
`'
`
`Fh = H cot vertical viewing angle
`2
`2
`
`Generate the Sphere-like Polyhedron
`
`Assumptions
`
`The following are four assumptions of our system:
`I. Geometric distortion of the camera lens is negligible
`(ideal perspective projection).
`2. The size, camera constants, horizontal and vertical
`view angles of different images are the same.
`3. All objects in the scene are fur enough from the
`camera such that the movement of the COP (center
`of projection) caused by panning the camera is
`negligible. Thus all optic axes of the images can be
`regarded as intersected at the COP.
`4. Each image is perpendicularly intersected by its
`optic axis at its center.
`
`Image Registration
`If we do not know the absolute orientation of the
`camera of each image, we can register the image by hand.
`In our implementation, we provide a tool for automatic
`registering by using the corresponding points within the
`related portion of the environment map. If the result is
`unacceptable, we provide another tool allowing a user to
`manually register images or modiJY the registration. The
`change can be processed in almost real -time.
`
`Model Generation
`After all images are registered, a texture mapped
`sphere-like polyhedron
`is generated. The essential
`concept of model generation is simple: generate a sphere(cid:173)
`like polyhedron with its textures generated by ray-casting
`on original images. The original images are arranged as
`textured polygons in the space by their registrations, as
`shown in the following figure.
`
`Polygl
`trapezo
`problen
`regular
`tbrougt
`rectang
`
`As fo1
`simply
`the trap
`of the o
`
`Gener.
`adjacen
`previou:
`solve th
`are ovet
`ray-cast
`of get c
`colors f
`colors '
`square
`corresp<
`below.
`
`where ~
`nearest 1
`
`By m
`disconti1
`smoothe
`
`Generate textured sphere-like polyhedron by ray-casting
`
`-398-
`
`

`

`the sphere-like polyhedron are all
`Polygons on
`trapezoids or triangles, so the storage of texture is a
`problem: non-regular images are harder to handle than
`regular ones. We simply divide the texture vertically
`through the center and re-combine the two halves into a
`rectangle as indicated above.
`
`As for determining the resolution of the texture, we
`simply let the resolution (pixel/degree) at the center of
`the trapezoid be the same as the resolution at the center
`of the original image.
`
`Smooth Intensity Discontinuity
`
`Generally, there are intensity discontinuity between
`adjacent images, so the environment map generated by
`previous ray-casting method will consist of patches. To
`solve the intensity discontinuity, it is suggested that there
`are overlapped portions between adjacent images, and the
`ray-casting method is modified for this purpose: instead
`of get color from the image first hit by the ray, we get
`colors from all images hit by the ray, and average these
`colors with the weight of each one. The weight is the
`the
`image) between
`(on
`the distance
`square of
`corresponding pixel and the nearest border, as indicated
`below.
`
`intensity difference, the overlapped portions may be
`insufficient. So the exposure should still be properly
`controlled to avoid large intensity difference between
`adjacent images when taking them.
`Sometimes the border of the images may have noise, so
`we can ignore colors of pixels too near to their nearest
`borders by using the weighted average ..
`
`Image Data Handling
`
`The data volume of images is generally tremendous. A
`512x512 digitized NTSC resolution image with R:G:B =
`take 0.5 MB
`5:6:5 bits (2 bytes per pixel) will
`(Megabytes). A typical sphere-like polyhedron made of
`about 90 such NTSC resolution images (with adjacent
`images overlapped) consists of 216 trapezoids will take
`about 21.2 MBI Thus the data transfer between main
`memory and storage device is relatively heavy, making
`image compression necessary not ouly for saving storage
`transfer. In our
`space but also for reducing data
`implementation, the JPEG image compression is applied
`to dramatically reduce the data size of textures. For the
`previous example polyhedron, the data size of texture is
`dramatically reduced from about 21.2MB to about 3MB,
`making it suitable for some large but relatively slow
`storing device such as CD-ROM, or for being transferred
`through existing network such as Internet. However, the
`is
`it
`time, so
`image decompression usually takes
`performed ouly when necessary (for images used by
`polygons that are going to be rendered right away) or
`when the CPU is idle. The priority to determine which
`image is decompressed first when CPU is idle or which
`decompressed image is released first when memory is not
`enough is simply by their relative location with respect to
`the viewing direction: the nearer image has higher
`priority.
`
`' c
`'•c
`1+ xl • 1
`result color= x.
`'
`'
`'
`Xt+Xz
`
`where x1, "2 are distances from the hit pixels to their
`nearest borders.
`
`intensity
`the
`By using such weighted average,
`discontinuity on the generated environment map will be
`images of large
`smoothed. However, for adjacent
`
`B
`
`viewing direction
`
`Priority: image B > image A: when CPU is idle, image B
`is decompressed before image A; when memory is
`insufficient, image A is released before image B.
`
`Object Viewer
`
`-399-
`
`

`

`The goal of "object viewer" is to let a user interactively
`observe an object from different views in real time. In
`other words, the system must 'show different views of an
`object. Thus, the essential idea of image-based method is
`that the system sequentially shows the images taken from
`the same object from different views, such as images A,
`B, and C in the following figure.
`
`A 2D prototype of object viewer
`
`There is one major problem: how many images should
`be taken? For full interaction, the images should be as
`many as possible, but this will use too much disk space.
`Thus we can just take a few images, and generate the
`intermediate frames from these reference images. The
`typical methods for the generation of the intermediate
`frames are computer vision techniques, such as multi(cid:173)
`view stereo and 3-D scene representation [6][7]. These
`methods generate reasonably good results, but they are
`time-consuming, and cannot be used
`in real-time
`applications. We find that a much simpler method is
`needed and described below. In short, the intermediate
`frame, say A', between A and B in the above figure is
`generated by interpolation. That is, the color of each
`pixel on A' is the weighted sum of the color of the
`corresponding pixel on A and B. Then we sequentially
`show images A, A', and then B.
`
`With this simple method, we find that for a 2-D
`implementation, that is, the reference views are arranged
`in a circle around the object in the horizontal plane, one
`image per 15 degrees may be enough. One intermediate
`frame is generated between two reference images. (Of
`course, we can also generate more than one intermediate
`frame, but one is enough.) The sequential show of the
`reference images and intermediate frames is good enough
`to fool human eyes. Thus the result would look like
`arbitrary rotation of an object.
`
`Result
`
`(size of 640 x 480 pixels, with R:G:B = 5:6:5 bits per
`pixel). These images are taken from the yard of the
`Agriculture college at National Taiwan University (-E;-;;1;:
`~~~ cp ~). The polyhedron consists of 216 trapezoids.
`By applying the JPEG image compression technique, the
`data size of the whole texture is dramatically reduced
`from 21.2MB to 3 MB. A walk through demo system is
`implemented on an IBM Pentium-133 PC (a 486-DX2 66
`is also suitable), under the MS-DOS operating system,
`using the Watcom C++ compiler (version IO.Oa), and an
`ET-4000 SuperVGA card. The Win95 version is still
`under development.
`
`A demonstration of the rendered results is given at the
`end of the paper. However, there is a practical problem
`remaining: it is almost impossible for some images to be
`registered without objects on the border being duplicated
`or lost in adjacent images because the camera was moved
`during panning.
`
`Frame rate measurement is shown below as a table and
`wes measured on
`a Pentium-133 PC.
`Image
`decompression time is not included, where
`rp
`is the
`latitude of viewing direction, as indicated in figure 5.
`The frame rate decreases while latitude increases because
`the number of trapezoids increases.
`
`Frame
`319xl99 639x479 319xl99 639x479
`size No double No double VESA
`VESA
`buffer
`buffer
`double
`double
`buffer
`buffer
`
`(D
`
`0
`15
`30
`45
`60
`65
`70
`75
`80
`85
`90
`
`44.31
`42.73
`41.14
`38.13
`35.26
`33.80
`32.76
`30.33
`27.64
`27.30
`25.28
`
`16.46
`16.38
`15.44
`14.66
`13.14
`12.60
`12.35
`11.83
`10.92
`10.57
`10.19
`
`34.76
`33.80
`32.99
`27.70
`23.33
`23.16
`22.74
`22.70
`19.85
`17.59
`16.99
`
`13.65
`13.00
`12.13
`11.94
`10.33
`10.27
`9.88
`9.10
`8.23
`8.09
`8.08
`
`The third and forth columns are the frame rates when
`we use the VESA standard to access ET4000 card and
`make one more page in video memory for double buffer.
`Double buffer provides better output quality, but there are
`additional overheads.
`
`one
`inte
`reference
`the frame
`Win95 by
`frame rat<
`
`Althougl
`method o
`does worl
`the illusio
`center of
`objects in
`to fo