\00 IEEE 6th Workshop on Multimedia Signal Processing
Low-delay Unconstrained Motion Compensated Temporal Filtering Technique for
Wavelet-based Fully Scalable Video Coding
Woo-Jin Han
DM R&D Center, Samsng Electronics
1, Maetan-dong, aldl-gu, -s, Gyeogg-do, 42-742, Kore
Telephone: +82-31-200-9319
Fa: +82-31-200-3427
Eail: whan.an@samsng.c
tt- Among many technologie use in wa velet-based
scalable ,·ideo coder, motion compensated temporal fltering
(MCT) is a key technolog for both reducing temporal reun
dancy and prol'iding fexible temporally scalable architecture.
Howe"er, since ƣ n-ebl e temporal pyramid structure in
trouce a a end-to-end delay of a size of GOP, it is "er
difcult to use scalable coding technolog many applications
that require end-to-end dla _ this paper, we po a
l-la unconstraine motion compensate temporal fltering
(D-UCTF) scheme. Our method fouse to control the end
to-end dea in a unifed framework el ha"e most useful
features of predicti"e coder, such multipl e reference frame
a ꭐ-a prediction. Experimental reults show that the
proposed method can pro"ide comparable cd efciency and
sigifcantly smaller performance ded compared to
the com'entional .
W
AVELET based scalable video coding has received
considerable attention fo man multimedia applica
tions in erms of boh coding aloriths and standardiaion
actives. Among an ecoloie used n wavelet-based
scalable vi deo coders, motion coenated emoral filteing
(MCTF) pposed by Ohm [1J and imped b Ch and
Woods [2] s a e ecolgy for both reducng teporal
redndanc and proiding feible temporally scalable archi
tectre.
More ecently, uconsranedMCTF(UMCTF) scheme was
rpsed by Taga and Mihaela [3] to ercoe seral
drabacks of oiinal MCTF scheme, such as ni-directional
motion esimaion. low qalit at lwe tempoal rates, and
increased dela. UCTF provides a more general fameo
enabe us o e mltple refeence frames and bi-directional
fltering. In the past researches, many epemental euts
hae shon that he UMCTF withot update-sep perfons
better than the oi ginl M. tis reao, the specc
form of UMCTF, which has no pdate-sep, is generall
sed [4] athough the most general f of UMCTF alos
an adaptie choice of ow-pa filters [5].
Many aplicaions that need low end-to-end dl ay, such as
ideo confeencng, rere low encoder-side dela as wel
as low dcode -side delay. Since both MCTF and UMCTP
11
HHHH, , I� --- ' �
L H l > L H I
G- �
L
& .
(0)
L "
2 [- �
t � L � l � L 1
ig. o o M. () Encoer-side.
() -.
(01 (0)
. 2. π remPra . () d-ξd.
() -.
analye frae from the lowet tmprl leels, he ecode
side deay shold be a sie of GOP, which is unacceptable for
ayeal-timeapcations. Alhogh canreduce the
dla euiemet by lmitng the umber of futre referece
rames, here iʦ no eplicit solution to corol the end-to-end
deay. Inthis paer, pp low-deayunconstrained mo
ion compensaed temporal prediction (LD-UMCTF). which is
basedon UMCTFalgorithm tocontroltheend-to-enddelayin
a nifed frameork haing most se features o UMCTF,
suc asmult i pe reeenceae and -dc predicton.
!
II. MC'F D UMCTP RMR
At st, we tar' from the dm of the tmoal pro
cessing fo in original MCTF. In Fig. , " fre" means a
lw rey or arage frame ad "H frame" means a hgh
fre-qenc or diference rame. As sonine gue, rame
p a the lowest: temporal eel ae processed irst n the
encode-side, whereas hose at he ighst evel ced
firs in the de-coder side.
,
0-7803-8578-0/04/$20.00 0 478