EE 5359

Project Report
Spring 2008
Study and Comparison of MPEG 2 and !2"# main
profi$es and a%ai$a&$e transcoding met'ods
Priyan(a )n(o$e(ar
*000 5* ##9+
,ist of )cronyms
AVC: Advanced video coding
CABAC: Context-based adaptive binary arithmetic coding
CAVLC: Context-based adaptive variable length coding
DCT: Discrete Cosine Transform
G!: Gro"p of pict"res
#DTV: #igh Definition Television
$DCT: $nverse DCT%
$&: $nverse '"anti(ation:
$): $nternational rgani(ation for )tandardi(ation%
$T*: $nternational telecomm"nication "nion%
+VT: +oint Video Team
,%-: ,otion -stimation
,%C: ,otion Compensation
,B: ,acrobloc.
,V: ,otion vector%
/AL: /et0or. abstraction layer
&!: &"anti(ation parameter%
VLC: Variable length coding
VLD: Variable length decoding
VCL: Video coding layer
VC-G: Video coding experts gro"p%
2
)&stract
There is a high demand for m"ltimedia applications li.e digital video recording
and teleconferencing% This has led to the development of vario"s video coding
standards li.e ,!-G-1 and #%123% The video coding layer of #%123 is
s"perficially similar to that of ,!-G-14 ho0ever5 there are several differences in
the details% $n this pro6ect the ,!-G-1 and #%123 video coding standards are
compared 0ith a concentration on the main profiles% #%123 gives a better
compression performance than ,!-G-1% #o0ever5 ,!-G-1 has already been
0idely "sed in the field of digital broadcasting5 #DTV and DVD applications% This
incompatibility problem bet0een #%123 video so"rce and the existing ,!-G-1
decoders can be solved "sing transcoders% This pro6ect also disc"sses the
criteria for efficient transcoding and a fe0 transcoding architect"res%
3
* -ntroduction
Development of the international video coding standards s"ch as ,!-G-1 789
7::97:897:29 boosted a diverse range of m"ltimedia applications5 incl"ding digital
video recording and teleconferencing% As a res"lt of the gro0ing demand for
better compression performance5 advanced standards s"ch as #%123 7:97197297;9
7:<9 0ere developed by the $T*-T-$)=$-C +oint Video Team >+VT? in 1@@A% The
overall scheme of the video coding layer >VCL? of #%123 is s"perficially similar to
the encoding scheme of ,!-G-1% #o0ever5 there are significant differences in
the details% $n this pro6ect the ,!-G-1 and #%123 video coding standards are
compared5 i%e% the similarities and differences are st"died5 0ith a concentration
on the main profiles%
#%123 can s"pport vario"s applications s"ch as video broadcasting5 video
streaming and video conferencing over fixed and 0ireless net0or.s and over
different transport protocols% #o0ever5 ,!-G-1 has already been 0idely "sed in
the field of digital broadcasting5 #DTV and DVD applications% The incompatibility
problem bet0een #%123 video so"rce and the existing ,!-G-1 decoders can be
solved by "sing transcoders% $n this pro6ect5 the criteria for transcoding and a fe0
transcoding architect"res are disc"ssed%
The report has been str"ct"red in the follo0ing manner: Chapter : is an
introd"ction to the topic and explains the scope of the pro6ect% Chapter 1 explains
the vario"s aspects of the ,!-G-1 video coding standard 0hile Chapter A
covers the same for #%123 video coding standard% Chapter 3 sho0s a
comparison bet0een the t0o standards% $n Chapter B5 the topic of ,!-G-1 to
#%123 transcoding is covered in greater detail%
4
2 MPEG.2
,!-G-1 is 0idely "sed as the format of digital television signals that are
broadcast by terrestrial >over-the-air?5 cable5 and direct broadcast satellite TV
systems% $t also specifies the format of movies and other programs that are
distrib"ted on DVD and similar dis.s% As s"ch5 TV stations5 TV receivers5 DVD
players5 and other e'"ipment are often designed to this standard% ,!-G-1 0as
the second of several standards developed by the ,oving !ict"res -xpert Gro"p
>,!-G? and is an international standard >$)=$-C :A<:<?% 7:29
The video section5 part 1 of ,!-G-15 is similar to the previo"s ,!-G-:
standard5 b"t also provides s"pport for interlaced video4 the format "sed by
analog broadcast TV systems% ,!-G-1 video is not optimi(ed for lo0 bit-rates5
especially less than : ,bit=s at standard definition resol"tions% #o0ever5 it
o"tperforms ,!-G-: at A ,bit=s and above% ,!-G-1 is directed at broadcast
formats at higher data rates of 3 ,bps >DVD? and :; ,bps >#DTV?% All
standards-compliant ,!-G-1 video decoders are f"lly capable of playing bac.
,!-G-: video streams% ,!-G-1=video is formally .no0n as $)=$-C :A<:<-1
and as $T*-T Cec% #%121 71:9%
2!* MPEG.2 Profi$es and ,e%e$s
,!-G-1 video s"pports 0ide range of applications from mobile to high '"ality
#D editing% Dor many applications5 it is "nrealistic and too expensive to s"pport
the entire standard% To allo0 s"ch applications to s"pport only s"bsets of it5 the
standard defines profile and level% 71:9
/escription
,!-G-1 video is a family of systems5 each having an arranged degree of
commonality and compatibility% $t allo0s fo"r so"rce formats5 or ELevelsF5 to be
coded5 ranging from Limited Definition >abo"t todayFs VCC3 '"ality?5 to f"ll
#DTVB G each 0ith a range of bit rates 7119% The level defines the s"bset of
'"antitative capabilities s"ch as maxim"m bit rate5 maxim"m frame si(e5 etc 7:29%
$n addition to this flexibility in so"rce formats5 ,!-G-1 allo0s different E!rofilesF%
-ach profile offers a collection of compression tools that together ma.e "p the
coding system% A different profile means that a different set of compression tools
is available% 7119
5
MPEG.2 Profi$es
2.1.1.1 Simple Profile
This profile has the fe0est tools% The )imple profile offers the basic tool.it for
,!-G-1 encoding% This is intra and predicted frame encoding and decoding 0ith
a color s"b sampling of H*V 3:1:@%
2.1.1.2 Main Profile
This profile has all the tools of the )imple !rofile pl"s one more >termed bi-
directional prediction?% $t gives better >maxim"m? '"ality for the same bit-rate than
the )imple !rofile% A ,ain !rofile decoder decodes both ,ain and )imple !rofile-
encoded pict"res% This bac.0ard compatibility pattern applies to the s"ccession
of profiles% A refinement of the ,ain !rofile5 sometimes "nofficially .no0n as
,ain !rofile !rofessional Level or ,!-G 3115 allo0s line-se'"ential color
difference signals >3:1:1? to be "sed5 b"t not the scaleable tools of the higher
!rofiles%
2.1.1.3 SNR Scalable Profile and Spatially Scalable Profile
The t0o !rofiles after the ,ain !rofile are5 s"ccessively5 the )/C )caleable
!rofile and the )patially )caleable !rofile% These add tools 0hich allo0 the
coded video data to be partitioned into a base layer and one or more Etop-"pF
signals% The top-"p signals can either improve the noise >)/C )calability? or the
resol"tion >)patial )calability?% These )caleable systems may have interesting
"ses% The lo0est layer can be coded in a more rob"st 0ay5 and th"s provide a
means to broadcast to a 0ider area5 or provide a service for more diffic"lt
reception conditions% /evertheless there 0ill be a premi"m to be paid for their
"se in receiver complexity% 0ing to the added complexity5 none of the )caleable
!rofiles is s"pported by digital video broadcasting >DVB?% The inp"ts to the
system are H*V component radio% #o0ever5 the first fo"r profiles code the color-
difference signals line-se'"entially%
2.1.1.4 High Profile
$t incl"des all the previo"s tools pl"s the ability to code line-sim"ltaneo"s colo"r-
difference signals% $n effect5 the #igh !rofile is a Es"per systemF5 designed for the
most sophisticated applications5 0here there is no constraint on bit rate%
Table : is a tab"lated form of the properties of the vario"s ,!-G-1 profiles%
6
Table :% ,!-G-1 !rofiles
MPEG.2 Profi$es0*"9
Abbr% /ame
!ict"re
Coding
Types
Chroma
Dormat
Aspect Catios )calable modes
SP )imple profile $5 ! 3:1:@
s'"are pixels5
3:A5 or :2:;
none
MP ,ain profile $5 !5 B 3:1:@
s'"are pixels5
3:A5 or :2:;
none
S1R
)/C )calable
profile
$5 !5 B 3:1:@
s'"are pixels5
3:A5 or :2:;
)/C >signal-to-
noise ratio?
scalable
Spatia$
)patially
)calable
profile
$5 !5 B 3:1:@
s'"are pixels5
3:A5 or :2:;
)/C- or spatial-
scalable
P #igh profile $5 !5 B
3:1:1 or
3:1:@
s'"are pixels5
3:A5 or :2:;
)/C- or spatial-
scalable
MPEG.2 ,e%e$s
7
2.1.1.5 Decription of !e"el
A level is the definition for the ,!-G standard for physical parameters s"ch as
bit rates5 pict"re si(es and resol"tions% There are fo"r levels specified by
,!-G1: #igh level5 #igh :33@5 ,ain level5 and Lo0 level% ,!-G-1 Video ,ain
!rofile and ,ain level has sampling limits at $T*-C 2@: parameters >!AL and
/T)C?% !rofiles limit syntax >i%e% algorithms? 0hereas Levels limit encoding
parameters >sample rates5 frame dimensions5 coded bitrates5 b"ffer si(e etc%?%
Together5 Video ,ain !rofile and ,ain Level >abbreviated as ,!I,L? .eep
complexity 0ithin c"rrent technical limits5 yet still meet the needs of the ma6ority
of applications% ,!I,L is the most 0idely accepted combination for most cable
and satellite systems4 ho0ever different combinations are possible to s"it other
applications% 739
Table 1 sho0s a comparison bet0een the fo"r ,!-G-1 levels on the basis of the
frame si(e >!AL=/T)C? and the maxim"m bit rate for each%
Table 1% ,!-G-1 Levels 7119
2!2 MPEG.2 Encoder
8
Dig"re :% ,!-G 1 encoder 7:@9
The vario"s bloc.s of the ,!-G-1 encoder are explained belo0:
/C2
The ,!-G-1 encoder "ses <x< 1-D DCT% $n the case of intra frames5 it is applied
to <x< bloc.s of pels and in the case of inter frames it is applied to <x< bloc.s of
the resid"al >motion compensated prediction errors?% )ince DCT is more efficient
in compressing correlated so"rces5 intra pict"res DCT compress more efficiently
than inter pict"res%
2!2!* 3uanti4er
The DCT coefficients obtained above are then '"anti(ed by "sing a defa"lt or
modified matrix% *ser defined matrices may be do0nloaded and can occ"r in the
se'"ence header or in the '"ant matrix extension header% The '"anti(er step
si(es for DC coefficients of the l"minance and chrominance components are <5 35
1 and : according to the intra DC precision of <5 ;5 :@ and :: bits respectively%
2!2!2 Motion estimation and compensation
,otion estimation and compensation: $n the motion estimation process5 motion
vectors for predicted and interpolated pict"res are coded differentially bet0een
macrobloc.s% The t0o motion vector components5 the hori(ontal component first
9
and then the vertical component are coded independently% The motion
compensation process forms prediction from previo"sly decoded pict"res "sing
the motion vectors that are of integer and half-pel resol"tions%
2!2!3 Coding decisions
There are fo"r different coding modes in ,!-G-1% These modes are chosen
based on 0hether the encoder encodes a frame pict"re as a frame or t0o fields
or in the case of interlaced pict"res it can chose to encode it as t0o fields or "se
:2x< motion compensation%
2!2!# Scanning and 5,C
The '"anti(ed transform coefficients are scanned and converted to a one
dimensional array% T0o scanning methods are available:
a% Jig(ag scan >Dig"re 1>a??: Dor progressive >non-interlaced? mode processing
b% Alternate scan >Dig"re 1>b??: Dor interlaced format video%
>a? >b?
Dig"re 1 >a? Jig Jag scan pattern >3x3? 739
>b? Alternate scan pattern >3x3?
10
>a?
>b?
Dig"re A% )can matrices in ,!-G-1 71@9 ><x<? >a? Jig(ag scan >b? Alternate scan
The list of val"es prod"ced by scanning is then entropy coded "sing a variable
length code >VLC?%
2!3 MPEG.2 /ecoder
11
Dig"re 3% ,!-G 1 Decoder 789
At the decoder side5 the '"anti(ed DCT coefficients are reconstr"cted and
inverse transformed to prod"ce the prediction error% This predicted error is then
added to the motion compensated prediction generated from previo"sly decoded
pict"re to prod"ce the reconstr"cted o"tp"t%
The vario"s parts of the ,!-G-1 decoder are:
2!3!* 5aria&$e $engt' decoding
This process involves the "se of a table defined for decoding intra DC
coefficients and three tables one each for non intra DC coefficients5 intra AC
coefficients and non intra AC coefficients% The decoded val"es basically infer one
of three co"rses of action: end of bloc.5 normal coefficients and escape coding%
2!3!2 -n%erse scan
The o"tp"t of the variable length decoding stage is one dimensional and of
length 23% $nverse scan process converts this one dimensional data into a t0o
dimensional array of coefficients according to a predefined scan matrix%
2!3!3 -n%erse 6uanti4ation
At this stage the t0o dimensional DCT coefficients are inverse '"anti(ed to
prod"ce the reconstr"cted DCT coefficients% This process involves the rescaling
of the coefficients by essentially m"ltiplying them by the '"anti(er step si(e% The
'"anti(er step si(e can be modified by "sing either a 0eighing matrix or a scale
12
factor% After performing inversion '"anti(ation5 sat"ration and mismatch control
operations are performed%
2!3!# -n%erse /C2
nce the reconstr"cted DCT coefficients are obtained5 a 1D <x< inverse DCT is
applied to obtain the inverse transformed val"es% These val"es are then
sat"rated to .eep them in the range of 7-1B2:K1BB9%
2!3!5 Motion Compensation
D"ring this stage5 predictions from previo"sly decoded pict"res are combined
0ith the inverse DCT transformed coefficient data to get the final decoded o"tp"t%
3 !2"#
13
#%123=AVC 7:97;9 0as developed by the +VT >+oint Video Team? to achieve
,!-G-1 789 '"ality compression at almost half the bit rate% #%123=AVC provides
significant coding efficiency5 simple syntax specifications5 and seamless
integration of video coding into all c"rrent protocols and m"ltiplex architect"res%
#%123 s"pports vario"s applications s"ch as video broadcasting5 video
streaming5 and video conferencing over fixed and 0ireless net0or.s and over
different transport protocols% 739
#%123 video coding standard has the same basic f"nctional elements as previo"s
standards >,!-G-:5 ,!-G-15 ,!-G-3 part 15 #%12:5 #%12A? 71A95 i%e%5 transform
for red"ction of spatial correlation5 '"anti(ation for bitrate control5 motion
compensated prediction for red"ction of temporal correlation5 entropy encoding
for red"ction of statistical correlation% #o0ever5 in order to f"lfill better coding
performance5 the important changes in #%123 occ"r in the details of each
f"nctional element by incl"ding intra-pict"re prediction5 a ne0 3x3 integer
transform5 m"ltiple reference pict"res5 variable bloc. si(es and a '"arter pel
precision for motion compensation5 a debloc.ing filter5 and improved entropy
coding% 7:9
3!* !2"# Profi$es
-ach !rofile specifies a s"bset of entire bitstream of syntax and limits that shall
be s"pported by all decoders conforming to that !rofile% There are three !rofiles
in the first version: Baseline5 ,ain5 and -xtended% Baseline !rofile is to be
applicable to real-time conversational services s"ch as video conferencing and
videophone% ,ain !rofile is designed for digital storage media and television
broadcasting% -xtended !rofile is aimed at m"ltimedia services over $nternet%
Also there are fo"r #igh !rofiles defined in the fidelity range extensions7:;9 for
applications s"ch as content-contrib"tion5 content-distrib"tion5 and st"dio editing
and post-processing : #igh5 #igh :@5 #igh 3:1:15 and #igh 3:3:3% #igh !rofile is to
s"pport the <-bit video 0ith 3:1:@ sampling for applications "sing high resol"tion%
#igh :@ !rofile is to s"pport the 3:1:@ sampling 0ith "p to :@ bits of
representation acc"racy per sample% #igh 3:1:1 !rofile is to s"pport "p to 3:1:1
chroma sampling and "p to :@ bits per sample% #igh 3:3:3 !rofile is to s"pport "p
to 3:3:3 chroma sampling5 "p to :1 bits per sample5 and integer resid"al color
transform for coding CGB signal% The !rofiles have both the common coding
parts and as 0ell specific coding parts as sho0n in Dig"re B% 7:9
3!*!*Common Parts of )$$ Profi$es
14
3.1.1.1 # lice $#ntra%coded lice&
This slice is coded by "sing prediction only from decoded samples 0ithin the
same slice%
3.1.1.2 P lice $Predicti"e%coded lice&
This slice >Dig"re 2? is coded by "sing inter prediction from previo"sly-decoded
reference pict"res5 "sing at most one motion vector and reference index to
predict the sample val"es of each bloc.%
3.1.1.3 '()!' $'onte*t%baed (dapti"e )ariable !ength 'oding&
This is "sed for entropy coding% After transform and '"anti(ation5 the probability
that the level of coefficients is (ero or K=-: is very high% CAVLC handles the (ero
and K=-: coefficients as the different manner 0ith the levels of coefficients% The
total n"mbers of (ero and K=-: are coded% Dor other coefficients5 their levels are
coded%
3!*!27ase$ine Profi$e
3.1.2.1 +le*ible macrobloc, order
,acrobloc.s may not necessarily be in the raster scan order% The map assigns
macrobloc.s to a slice gro"p%
3.1.2.2 (rbitrary lice order
The macrobloc. address of the first macrobloc. of a slice of a pict"re may be
smaller than the macrobloc. address of the first macrobloc. of some other
preceding slice of the same coded pict"re%
3.1.2.3 Red-ndant lice
This slice belongs to the red"ndant coded data obtained by same or different
coding rate5 in comparison 0ith previo"s coded data of same slice%
15
Dig"re B% The specific coding parts of the !rofiles in #%123 7:9%
3!*!3 Main Profi$e
3.1.3.1 . lice $.i%directionally predicti"e%coded lice&
This slice >Dig"re 2? is coded by "sing inter prediction from previo"sly-decoded
reference pict"res5 "sing at most t0o motion vectors and reference indices to
predict the sample val"es of each bloc.%
3.1.3.2 /eighted prediction
This is a scaling operation performed by applying a 0eighting factor to the
samples of motion-compensated prediction data in ! or B slice% A prediction
signal p for B slice is obtained "sing different 0eights from t0o reference signals5
r: and r1%
-'"ation : 7:9: p L 0: r: K 01 r1
0here 0: and 01 are 0eights%
16
3.1.3.3 '(.(' $'onte*t%baed (dapti"e .inary (rithmetic 'oding&
This is "sed for entropy coding% $t "tili(es arithmetic coding in order to achieve
good compression%
Dig"re 2% $ll"stration of temporal prediction >B and ! slices?
3!*!# E8tended Profi$e
This profile incl"des all parts of Baseline !rofile: flexible macrobloc. order5
arbitrary slice order5 and red"ndant slice% The other feat"res of this profile are:
3.1.4.1 SP lice
The specially coded slice for efficient s0itching bet0een video streams5 similar to
coding of a ! slice%
3.1.4.2 S# lice
The s0itched slice5 similar to coding of an $ slice%
3.1.4.3 Data partition
The coded data is placed in separate data partitions5 each partition can be
placed in different layer "nit%
17
3.1.4.4 . lice
#%123 generali(es the concept of bidirectional prediction and s"pports not only
for0ard=bac.0ard prediction pairs b"t also for0ard=for0ard and
bac.0ard=bac.0ard pairs%
3.1.4.5 /eighted prediction
All existing standards consider e'"al 0eights for reference pict"res5 i%e% a
prediction signal is obtained by averaging "sing e'"al 0eights of reference
signals% B"t grad"al transitions from scene to scene need different 0eights%
#%123 "ses 0eighted prediction method for a macrobloc. of ! slice or B slice%
3!*!5 ig' Profi$es
#igh profiles incl"de all parts of the ,ain !rofile: B slice5 0eighted prediction5
CABAC% The salient feat"res of this profile are:
3.1.5.1 (dapti"e tranform bloc, i1e
#%123 "ses an adaptive transform bloc. si(e5 3 x 3 and < x < >#igh !rofiles only?5
0hereas previo"s video coding standards "sed the < x < DCT% The smaller bloc.
si(e leads to a significant red"ction in ringing artifacts% Also5 the 3 x 3 transform
has the additional benefit of removing the need for m"ltiplications% 7:9
3.1.5.2 2-anti1ation caling matrice
Different scaling according to specific fre'"ency associated 0ith the transform
coefficients in the '"anti(ation process to optimi(e the s"b6ective '"ality% The
#igh !rofiles s"pport the percept"al-based '"anti(ation scaling matrices similar
to those "sed in ,!-G-1% The encoder can specify a matrix for scaling factor
according to the specific fre'"ency associated 0ith the transform coefficient for
"se in inverse '"anti(ation scaling by the decoder% This allo0s optimi(ation of the
s"b6ective '"ality according to the sensitivity of the h"man vis"al system 0hich is
less sensitive to the coded error in high fre'"ency transform coefficients%
Table A sho0s a comparison bet0een the baseline5 extended5 main and high
profiles of #%123%
18
Table A% Comparison chart for the vario"s profiles of #%123 7:<9
7ase$ine E8tended Main ig'
$ and ! )lices Hes Hes Hes Hes
B )lices /o Hes Hes Hes
)$ and )! )lices /o Hes /o /o
,"ltiple Ceference Drames Hes Hes Hes Hes
$n-Loop Debloc.ing Dilter Hes Hes Hes Hes
CAVLC -ntropy Coding Hes Hes Hes Hes
CABAC -ntropy Coding /o /o Hes Hes
Dlexible ,acrobloc. rdering >D,? Hes Hes /o /o
Arbitrary )lice rdering >A)? Hes Hes /o /o
Ced"ndant )lices >C)? Hes Hes /o /o
19
3!2 !2"# Encoder
Dig"re 8% #%123 encoder 7;9
The encoder bloc.s are explained belo0:
3.2.1.1 4*4 #nteger tranform
The #%123 employs a 3x3 integer DCT as compared to <x< DCT adopted by the
previo"s standards% The smaller bloc. si(e leads to a significant red"ction in
ringing artifacts% Also5 the 3 x 3 transform has the additional benefit of removing
the need for m"ltiplications%
3.2.1.2 2-anti1ation and can
The #%123 standard specifies the mathematical form"lae of the '"anti(ation
process% The scale factor for each element in each s"b bloc. varies as a f"nction
of the '"anti(ation parameter associated 0ith the macrobloc. and as a f"nction
20
of the position of the element 0ithin the s"b bloc.% The rate control algorithm
controls the val"e of the '"anti(ation parameter% T0o types of scan pattern are
"sed for 3x3 bloc.s G one for frame coded macrobloc.s and one for field coded
macrobloc.s%
3.2.1.3 'onte*t%baed adapti"e "ariable length coding $'()!'& and 'onte*t%
baed adapti"e binary arithmetic coding $'(.('& entropy coding
#%123 "ses different variable length coding methods in order to match a symbol
to a code based on the context characteristics% They are context-based adaptive
variable length coding >CAVLC? and context-based adaptive binary arithmetic
coding >CABAC?% All syntax elements except for the resid"al data are encoded by
the -xp-Golomb codes% $n order to read the resid"al data >'"anti(ed transform
coefficients?5 (ig-(ag scan >interlaced? or alternate scan >non-interlaced or field?
is "sed% Dor coding the resid"al data5 a more sophistical method called CAVLC is
employed% Also5 CABAC is employed in ,ain and #igh profiles5 CABAC has
more coding efficiency b"t higher complexity compared to CAVLC%
3.2.1.4 Debloc,ing filter
#%123 employs a debloc.ing filter to red"ce the bloc.ing artifacts in the bloc.
bo"ndaries and stops the propagation of acc"m"lated coded noise% The filter is
applied after the inverse transform >before reconstr"cting and storing the
macrobloc. for f"t"re predictions? and in the decoder >before reconstr"cting and
displaying the macrobloc.s?% The debloc.ing filter is applied across the edges of
the macrobloc.s and the s"b-bloc.s% The filtered image is "sed in motion
compensated prediction of f"t"re frames and helps achieve more compression%
Dig"re <% Diagram depicting ho0 the loop filter 0or.s on the edges of the bloc.s
and s"b-bloc.s 739
21
3.2.1.5 #ntra prediction
D"ring intra prediction5 the encoder derives a predicted bloc. based on its
prediction 0ith previo"sly decoded samples% The predicted bloc. is then
s"btracted from the c"rrent bloc. and then encoded% There are a total of nine
prediction modes >Dig"re ;? for each 3x3 l"ma bloc.5 fo"r prediction modes for
each :2x:2 l"ma bloc. and fo"r modes for each chroma bloc.%
Dig"re ;% $ntra prediction 3x3 7A:9
3.2.1.3 #nter prediction
$nter prediction is performed on the basis of temporal correlation and consists of
motion estimation and motion compensation% As compared to the previo"s
standards5 #%123 s"pports a large n"mber of bloc. si(es from :2x:2 to 3x3%
,oreover #%123 s"pports motion vector acc"racy of one-'"arter of the l"ma
sample%
3.2.1.4 Reference pict-re
*nli.e the previo"s standards that 6"st "se the immediate previo"s $ or ! pict"re
for inter prediction5 #%123 has the ability to "se more than one previo"s reference
pict"re for inter prediction th"s enabling the encoder to search for the best match
for the c"rrent pict"re from a 0ider set of reference pict"res than 6"st the
previo"sly encoded one%
22
3!3 !2"# /ecoder
Dig"re :@ sho0s the bloc. diagram of a general #%123=,!-G-3 AVC decoder%
Dig"re :@% #%123 decoder 789
$t incl"des all the control information s"ch as pict"re or slice type5 macrobloc.
types and s"btypes5 reference frames index5 motion vectors5 loop filter control5
'"anti(er step si(e etc5 as 0ell as coded data comprising of '"anti(ed transform
coefficients% The decoder of Dig"re :@ 0or.s similar to the local decoder at the
encoder4 a simplified description is as follo0s% After entropy >CABAC or CAVLC?
decoding5 the transform coefficients are inverse scanned and inverse '"anti(ed
prior to being inverse transformed% To the res"lting 3M3 bloc.s of resid"al signal5
an appropriate prediction signal >intra or motion compensated inter? is added
depending on the macrobloc. type mbtyp >and s"bmbtype? mode5 the reference
frame5 the motion vector=s5 and decoded pict"res store5 or in intra mode% The
reconstr"cted video frames "ndergo debloc. filtering prior to being stored for
f"t"re "se for prediction% The frames at the o"tp"t of debloc.ing filter may need
to "ndergo reordering prior to display% [29
23
# Comparison &et9een MPEG.2 and !2"#
#!* :ey features of MPEG.2 %ideo
The ,!-G-1 coding standard has been designed to efficiently s"pport both
interlaced and progressive video coding and prod"ce high '"ality standard
definition video at abo"t 3 ,bps% The ,!-G-1 video standard "ses a bloc.-
based hybrid transform coding algorithm that employs transform coding of the
motion-compensated prediction error% Nhile motion compensation exploits
temporal red"ndancies5 the DCT transform exploits the spatial red"ndancies% The
asymmetric encoder-decoder complexity allo0s for a simpler decoder 0hile
maintaining high '"ality and efficiency thro"gh a more complex encoder% 7A9
#!2 :ey features of !2"# %ideo
The #%123 video coding standard has been developed recently thro"gh the 6oint
0or. of the $T*Fs video coding experts gro"p >VC-G? and $) moving pict"res
experts gro"p >,!-G?% The #%123 video coding standard is flexible and offers a
n"mber of tools to s"pport a range of applications 0ith very lo0 as 0ell as very
high bitrate re'"irements%
#!3 Comparison; Simi$arities and /ifferences &et9een MPEG.2 %ideo
and !2"# %ideo
$n this section5 the ,!-G-1 and #%123 video coding standards are compared
0ith respect to their vario"s aspects s"ch as bit rate5 bloc. si(e5 macrobloc. si(e5
intra prediction5 motion estimation bloc.s5 '"anti(ation5 motion vector prediction5
intra prediction amongst vario"s other% Table 3 tab"lates these comparisons
systematically% They are also f"rther elaborated in the s"b sections belo0%
#!3!* -ncreased efficiency
Compared 0ith ,!-G-1 video5 the #%123 video format gives percept"ally
e'"ivalent video at :=A to:=1 of the ,!-G-1 bit rates% )ome extensions--.no0n
as ODidelity Cange -xtensionsO facilitate higher-fidelity video coding by s"pporting
higher bit-depths5 incl"ding :@-bit and :1-bit encoding5 and higher color
resol"tion "sing the sampling str"ct"res H*V 3:1:1 and H*V 3:3:3% This nat"rally
ma.es it attractive to video distrib"tors5 beca"se it permits them to maximi(e the
n"mber of services that may be contained in a given amo"nt of band0idth 7A@9%
The bit rate gains are not a res"lt of any single feat"re b"t a combination of a
n"mber of encoding tools% These gains come 0ith a significant increase in
24
encoding and decoding complexity 7189% $n spite of the increased complexity5 the
dramatic band0idth savings enco"rages TV broadcasters to adopt the ne0
technology as they can "se the band0idth savings to provide ne0 channels or
ne0 data and interactive services% Nith the coding gains of #%1235 f"ll length
#DTV resol"tion movies can be stored on DVDs% D"rther more5 the fact that the
same video coding format can be "sed to broadcast TV as 0ell as for internet
streaming%
#!3!2 Coding f$e8i&i$ity
$):33;2-:@=#%1235 li.e previo"s ,!-G standards5 does not define a specific
encoder and decoder% $nstead5 it defines the syntax of an encoded bitstream and
describes the method of decoding that bitstream% The implementation is left to
the developer% 7A:9
The #%123 video "ses the same hybrid coding approach that is "sed in the other
,!-G video standards: motion compensated transform coding% The #%123
employs a hybrid coding approach similar to that of ,!-G-1 b"t differs
significantly from ,!-G-1 in terms of the act"al coding tools "sed% The main
differences are: "se of an integer transform 0ith energy compaction properties
similar to that of the DCT instead of the DCT5 an in-loop debloc.ing filter >DD? to
red"ce bloc. artifacts5 and intra frame prediction >$D!?% The coder control
operation is responsible for f"nctions s"ch as reference frame management5
coding mode selection5 and managing the encoding parameter set% Besides5 the
#%123 standard introd"ces several other ne0 coding tools that improve coding
efficiency%
,"ltiple reference pict"re motion compensation "ses previo"sly encoded
pict"res more flexibly than does ,!-G-1% $n ,!-G-15 a !-frame can "se only a
single previo"sly coded frame to predict the motion compensation val"es for an
incoming pict"re5 0hile a B-frame can "se only the immediately previo"s !- or $-
frame and the immediately s"bse'"ent !- or $-frame%
#%123 permits the "se of "p to A1 previo"sly coded pict"res5 and it s"pports
more flexibility in the selection of motion compensation bloc. si(es and shapes5
do0n to the "se of a l"ma compensation bloc. as small as 3-by-3 pixels% #%123
also s"pports '"arter-sample motion compensation vector acc"racy5 as opposed
to ,!-G-1Ps half-sample acc"racy%
These refinements permit more precise segmentation of moving areas 0ithin the
image5 and more precise description of movement% D"rther5 in #%1235 the motion-
compensated prediction signal may be 0eighted and offset by the encoder5
25
facilitating significantly improved performance in fades >fades can be problematic
for ,!-G-1?%
#!3!3 /e&$oc(ing fi$ter
Bloc.-based coding can generate bloc.ing artifacts in the decoded pict"res% $n
#%1235 a de-bloc.ing filter is bro"ght 0ithin the motion-compensated prediction
loop5 so that this filtering may be "sed to predict an expanded n"mber of pict"res
>Dig"re <?%
)0itching slices5 0hich permit a decoder to 6"mp bet0een bitstreams in order to
smoothly change bit-rates or do st"nt modes 0itho"t re'"iring all streams to
send an $-frame at the s0itch point >ma.ing the decoderPs 6ob easier at s0itch
points?5 have been incorporated%
Table 3% Comparison bet0een ,!-G-1 and #%123
)$gorit'm C'aracteristic MPEG.2 !2"#
26
General ,otion
compensated
predictive5 resid"al
transformed5
entropy coded
)ame basic str"ct"re as
,!-G
Bloc. si(e <x<
:2x:25 <x:25 :2x<5 <x<5
3x<5 <x35 3x3
,acrobloc. si(e :2x:2 >frame mode?
:2x< >field mode?
:2x:2
$ntra !rediction /one ,"lti-direction5 ,"lti-
pattern
&"anti(ation
)calar '"anti(ation
0ith step si(e of
constant increment
)calar '"anti(ation 0ith
step si(e of increase at the
rate of :1%BQ
-ntropy coding VLC CAVLC5 CABAC
Neighted prediction /o Hes
Ceference pict"re ne pict"re ,"ltiple pict"res
,otion -stimation Bloc.s :2x:2 :2x:25 <x:25 :2x<5 <x<5
3x<5 <x35 3x3
,otion vector prediction )imple *ses median and
segmented
-ntropy Coding ,"ltiple VLC Tables Arithmetic Coding and
adaptive VLC Tables
Drame Distance for
!rediction
K=- : *nlimited
for0ard=bac.0ard
Dractional ,otion
-stimation
:=1 !ixel :=3 !ixel
Debloc.ing Dilter /one Dynamic edge filters
)calable coding s"pport 719 Hes5 layered pict"re
spatial5 )/C5
temporal scalability
Nith some s"pport on
temporal and )/C
scalability
Bit rates 0ith same '"ality
#D video 0ith resol"tion
>:;1@ x :@<@?
:1 -1@ ,bps 8 G < ,bps
Transmission rate 1 G :B ,bps 23 .bps G :B@ ,bps
#!3!# Performance comparison &et9een MPEG.2 and !2"# using standard
test streams < Simu$ation resu$ts
27
Test streams >foreman5 ne0s and carphone 7129? 0ere encoded "sing the open-
so"rce ,!-G-1 codec 71B9 and the #%123 codec 7139% The res"lts 0ere
compared against each other for parameters li.e the signal to noise ratio >)/C?5
G! and compression ratio% C$D files 0ere "sed for the RDoremanS and the
R/e0sS clips 0hereas &C$D 0as "sed for the RCarphoneS clip% The bit rate for
#%123 encoding 0as ta.en as the standard one "sed by the codes% The bit rate
for ,!-G-1 encoding 0as ad6"sted on the basis of the bit rate of the #%123
encoding process% This helped to compare the t0o standards on a common
plane% Nhile the aim of this pro6ect is to compare the ,ain profiles of ,!-G-1
and #%1235 sim"lations 0ere r"n for the )imple=Baseline profiles too% This 0as
done in order to prove '"antitatively that encoding "sing the ,ain profile for both
,!-G-1 and #%123 gives a better compression ratio and better '"ality video
than the )imple profile% Tables B5 2 and 8 tab"late the res"lts obtained after
r"nning the sim"lations% Dig"res ::5 :1 and :A sho0 screen shots of the encoded
videos >only for the ,ain profiles?% )ection 3%A%B explains the concl"sions dra0n
on the basis of the res"lts obtained from sim"lations%
#!3!5 Conc$usion
Drom the tables belo05 the follo0ing is concl"ded:
Dor the same bit rate and video resol"tion5 the !)/C >dB? val"es are
greater for #%123 encoded videos than for the ,!-G-1 encoded videos
indicating better video '"ality% This can be verified from the screen shots%
The compression ratio for #%123 encoded video is also better than that for
,!-G-1 encoded video inspite of better '"ality video%
Compression ratio L original file si(e=compressed file si(e
The video '"ality for #%123 video is better than for ,!-G-1 video for the
)imple=Baseline profiles as 0ell% Therefore5 it can be concl"ded that #%123
video coding standard gives better compression and better video '"ality
as compared to ,!-G-1%
Table B% !erformance comparison bet0een ,!-G-1 and #%123 main=simple
profiles - Doreman
28
Parameter Main Profi$e Simp$e profi$e
MPEG.2 !2"# MPEG.2 !2"#
$np"t video
resol"tion
AB1 x 1<<
>C$D?
AB1 x 1<<
>C$D?
AB1 x 1<<
>C$D?
AB1 x 1<<
>C$D?
fps A@ A@ A@ A@
T frames
encoded
;@ ;@ ;@ ;@
G! $-!-B-B-!-
B-B
$-B-B-!-B-
B-!
$-!-!-! $-!-!-!
!)/C >H? >dB? A@%31 A8%@A A1%: A8%3
!)/C >*? >dB? A;%:
3:%@< A;%@: 3:
!)/C >V? >dB? A;%2 3A%<: 3@%1 3A%2
Bit rate
>.bits=second?
3<:%@@ 3<:%@2 B2:%@ B2:%@:
Compression
ratio
83:: 8<:: 2B:: 2B::
>a? >b?
Dig"re ::% >a? Doreman G ,!-G-1 >main profile? encoding
>b? Doreman G #%123 >main profile? encoding
Table 2% !erformance comparison bet0een ,!-G-1 and #%123 main profiles G
/e0s
29
Parameter Main Profi$e Simp$e Profi$e
MPEG.2 !2"# MPEG.2 !2"#
$np"t video
resol"tion
AB1 x 1<<
>C$D?
AB1 x
1<< >C$D?
AB1 x
1<< >C$D?
AB1 x 1<<
>C$D?
fps A@ A@ A@ A@
T frames
encoded
;@ ;@ ;@ ;@
G! $-!-B-B-!-
B-B
$-B-B-!-
B-B-!
$-!-!-! $-!-!-!
!)/C >H? >dB? A8%@1 A;%: A3%@: A;
!)/C >*? >dB? A8%@1
3:%@ A;%: 3:
!)/C >V? >dB? A;%@1 31%@ A;%8 31
Bit rate
>.bits=second?
A82%@@ A82%@@ A<@%< A<@%<
Compression
ratio
;3:: ;;%8:: ;B%B:: ;B%B::
>a? >b?
Dig"re :1%>a? /e0s G ,!-G-1 >main profile? encoding
>b? /e0s G #%123 >main profile? encoding
Table 8% !erformance comparison bet0een ,!-G-1 and #%123 main profiles G
Carphone
Parameter Main Profi$e Simp$e Profi$e
MPEG.2 !2"# MPEG.2 !2"#
30
$np"t video
resol"tion
:82 x :33
>&C$D?
:82 x
:33
>&C$D?
:82 x
:33
>&C$D?
:82 x :33
>&C$D?
fps A@ A@ A@ A@
T frames
encoded
;@ ;@ ;@ ;@
G! $-!-B-B-!-
B-B
$-B-B-!-
B-B-!
$-!-!-! $-!-!-!
!)/C >H? >dB? A@%32 A8%2 A:%2 A<
!)/C >*? >dB? A2%A2
3@%; A; 3@%<
!)/C >V? >dB? A2%B 3:%B A; 3:%A
Bit rate
>.bits=second?
:1< :18%2 :38%A :38%A
Compression
ratio
2;%2:: 81%2:: 2@%<:: 2:%;::
>a? >b?
Dig"re :A% >a? Carphone G ,!-G-1 >main profile? encoding
>b? Carphone G #%123 >main profile? encoding
5 2ranscoding met'ods
5!* -ntroduction to transcoding
$n this fast gro0ing 0orld of m"ltimedia and telecomm"nications there is a great
demand for efficient "sage of the available band0idth% Nith the gro0th of
31
technology there is an increase in the n"mber of net0or.s5 types of devices and
different content representation formats as a res"lt of 0hich interoperability
bet0een different systems and net0or.s is gaining in importance% Transcoding of
video content is one s"ch effort in this direction% Besides these5 a transcoder can
also be "sed to insert ne0 information for example companyFs logos5 0atermar.s
as 0ell as error resilience feat"res into a compressed video stream% Transcoding
techni'"es are also "sef"l in s"pporting VCC tric. modes s"ch as fast-for0ard5
reverse play etc% for on-demand applications% 739
Technically5 transcoding is the coding and recoding of digital content from one
compressed format to another to enable transmission over different media and
playbac. over vario"s devices 71;9%
#aving said this5 no0 arises the '"estion of 0hy the need for #123=AVC to
,!-G-1 transcoding 7:39 7:B9U $n order to provide better compression of video
as compared to previo"s standards5 #%123=AVC 0as recently developed by the
+VT >+oint Video Team?% This ne0 standard f"lfills significant coding efficiency5
simple syntax specifications and seamless integration of video coding into all
c"rrent protocols and m"ltiplex architect"res% The #%123 specification represents
a significant advancement in the field of video coding technology by providing
,!-G-1 comparable video '"ality at an average of half the re'"ired band0idth%
)ince 0idespread "se of #%123 is anticipated5 many legacy systems incl"ding all
Digital TVs and home receivers "se ,!-G-1% This leads to the need for an
efficient architect"re that significantly employs the lo0er cost of #%123 video and
does not re'"ire a significant investment in additional video coding hard0are%
Dig"re :3% #%123 to ,!-G-1 transcoder applications 7:19
5!2 o9 is transcoding done < t'e &asic process
The simplest approach to transcoding is to completely decode the ,!-G-1 bit
stream and then re-encode it 0ith an #%123 encoder% The decode operation can
32
be performed either externally or as a part of the #%123 encoder% )ystem iss"es5
s"ch as handling )CT--AB digital program insertion >D!$? messages5 0ill re'"ire
that the decode and the encode operations be tightly co"pled% The '"ality of
transcoding 0ith this simple approach 0ill not be high%
Dig"re :B sho0s a comparison bet0een direct encoding and transcoding% The
fig"re sho0s the !)/C >a meas"re of mean s'"are error bet0een the inp"t and
decoded o"tp"t? val"es comp"ted at different bit rates% The !)/C n"mbers are
obtained by averaging the res"lts over :< different se'"ences of varying content
type and complexities% The top plot sho0s the performance of direct encoding
"sing an #%123 encoder% The bottom plot sho0s the performance of transcoding
0here the video is originally coded 0ith ,!-G-1 at 3,b=s5 decoded and then re-
encoded 0ith the same encoder "sed for direct encoding% Transcoding can res"lt
in "p to 1@ percent loss in compression efficiency%
)imilar to the previo"s approach5 the incoming ,!-G-1 stream is decoded and
then re-encoded "sing an #%123 encoder% #o0ever5 here the relevant information
available from the ,!-G-1 bit stream is re"sed%
Dig"re :B% !erformance comparison bet0een direct encoding and transcoding
7A19
5!3 Criteria for transcoding
33
Transcoding can be of vario"s types 7:39% )ome of them are bit rate transcoding
to facilitate more efficient transport of video5 spatial and temporal resol"tion
red"ction transcoding for "se in mobile devices 0ith limited display and
processing po0er and error-resilience transcoding in order to achieve higher
resilience of the original bit stream to transmission errors%
To achieve optim"m res"lts by transcoding5 the follo0ing criteria have to be
f"lfilled:
>i? The '"ality of the transcoded bitstream sho"ld be comparable to the one
obtained by direct decoding and re-encoding of the o"tp"t stream%
>ii? The information contained in the inp"t stream sho"ld be "sed as m"ch as
possible to avoid m"ltigenerational deterioration%
>iii? The process sho"ld be cost efficient5 lo0 in complexity and achieve the
highest '"ality possible%
5!# 2ranscoding of !2"# to MPEG.2
$n order to provide better compression of video as compared to previo"s
standards5 #%123=AVC video coding standard 0as recently developed by the +VT
>+oint Video Team? consisting of experts from VC-G >Video Coding -xperts
Gro"p? and ,!-G% This ne0 standard f"lfills significant coding efficiency5 simple
syntax specifications5 and seamless integration of video coding into all c"rrent
protocols and m"ltiplex architect"res% Th"s #%123 can s"pport vario"s
applications s"ch as video broadcasting5 video streaming and video conferencing
over fixed and 0ireless net0or.s and over different transport protocols% #o0ever
,!-G-1 has already been 0idely "sed in the field of digital broadcasting5 #DTV
and DVD applications% #ence transcoding is a feasible method to solve the
incompatibility problem bet0een #%123 video so"rce and the existing ,!-G-1
decoders%
An #%123=AVC to ,!-G-1 transcoder is designed to transcode the #%123 video
stream to ,!-G-1 format so as to be "sed by the ,!-G-1 end e'"ipment% $t is
better to transmit #%123 bitstreams on p"blic net0or.s to save on the m"ch
needed band0idth and then transcode them into ,!-G-1 bitstreams for local
,!-G-1 e'"ipment li.e a set-top box%
5!5 2ranscoding arc'itectures
This section describes the vario"s transcoding architect"res 7:B9:
34
5!5!* =pen $oop transcoding;
pen loop transcoders incl"de selective transmission 0here the high fre'"ency
DCT coefficients are discarded and re'"anti(ation% They are comp"tationally
efficient5 since they operate directly on the DCT coefficients% #o0ever they s"ffer
from the drift problem% Drift error occ"rs d"e to ro"nding5 '"anti(ation loss and
clipping f"nctions%
Dig"re :2% pen loop transcoding architect"re 7:B9
5!5!2 Cascaded pi8e$ domain transcoding arc'itecture;
This is a drift free architect"re% $t is a concatenation of a simplified decoder and
encoder as sho0n in Dig"re :8% $n this architect"re5 instead of performing the f"ll
motion estimation5 the encoder re"ses the motion vectors along 0ith other
information extracted from the inp"t video bitstream th"s red"cing the complexity%
Dig"re :8% Cascaded pixel domain transcoding architect"re 7:B9%
Simp$ified /C2 domain transcoding >S//2?;
This architect"re is based on the ass"mption that DCT5 $DCT and motion
compensation are all linear operations% )ince in this architect"re5 the motion
compensation is performed in the DCT domain it is a comp"tationally intensive
35
operation% Dor instance5 as sho0n in Dig"re :;5 the goal is to comp"te the target
bloc. B from the fo"r overlapping bloc.s B:5 B15 BA and B3%
Dig"re :<% )implified DCT domain transcoding architect"re 7:B9%
Dig"re :;% DCT- ,otion compensation 7:B9%

)DDT eliminates the DCT=$DCT and red"ces the frame n"mbers by half as a
res"lt of 0hich it re'"ires less comp"tation and memory as compared to C!DT%
#o0ever the linearity ass"mptions are not strictly tr"e since there are clipping
f"nctions performed in the video encoder=decoder and ro"nding operations
performed in the interpolation for fractional pixel ,C% These failed ass"mptions
may ca"se drift in the transcoded video%
5!5!3 Cascaded /C2 domain transcoding >C//2?
The cascaded DCT-domain transcoder can be "sed for spatial and temporal
resol"tion do0nscaling and other coding parameter changes% Compared to
)DDT5 greater flexibility is achieved "sing additional DCT-motion compensation
36
and frame memory res"lting in higher cost and complexity% This architect"re is
adopted for do0nscaling operations 0here the encoder side DCT-,C and
memory 0ill not cost m"ch%
Dig"re 1@% Cascaded DCT domain transcoding architect"re 7:B9
5!" Conc$usions
The selection of appropriate transcoding architect"re depends "pon the
application for 0hich it is intended% There is generally a tradeoff bet0een the
acc"racy and the complexity and cost of the architect"re% Dor example5 the
simplest open loop architect"re is the easiest to implement b"t it s"ffers from the
problem of drift 0hereas the cascaded DCT domain transcoding architect"re
overcomes this problem b"t it is a very complex and expensive architect"re to
implement%
" References
7:9 )oon-.a. V0on5 A% Tamhan.ar and V%C% Cao5 Rvervie0 of #%123 = ,!-G-3
!art :@ >pp%:<2-1:2?S5 )pecial iss"e on R -merging #%123=AVC video coding
standardS5 +% Vis"al Comm"nication and $mage Cepresentation5 vol% :85 pp%:<A-
BB15 Apr% 1@@2%
37
719 A% !"ri5 #% Chen and A% L"thra5 RVideo Coding "sing the #%123=,!-G-3 AVC
compression standardS5 )ignal !rocessing: $mage Comm"nication5 vol%:;5 pp
8;A-<3;5 ct% 1@@3%
7A9 #% Valva5 R$ss"es in #%123=,!-G-1 Video TranscodingS5 Comp"ter )cience
and -ngineering5 Dlorida Atlantic *niversity5 Boca Caton5 DL%
739 )% )harma5 RTranscoding of #%123 bitstream to ,!-G 1 bitstreamS5 ,asterFs
Thesis5 ,ay 1@@25 -- Department5 *niversity of Texas at Arlington%
7B9 )% )harma and V% C% Cao5 RTranscoding of #%123 bitstream to ,!-G-1
bitstreamS5 !roceedings of Asia-!acific Conference on Comm"nications 1@@8%
729 R-merging #%123=AVC Video Coding )tandardS5 +% Vis"al Comm"nication and
$mage Cepresentation5 vol%:85 pp% :<A-BB15 Apr% 1@@2%
789 !%/%T"dor5 RT"torial on ,!-G-1 Video CompressionS5 $-- + Langham
Thomson !ri(e5 -lectronics and Comm"nication -ngineering +o"rnal5 Dec% :;;B%
7<9 RThe ,!-G-1 $nternational )tandardS5 $)=$-C5 Ceference n"mber $)=$-C
:A<:<-15 :;;2%
7;9 T% Niegand et% al%5 Rvervie0 of the #%123=AVC Video Coding )tandardS5 $---
Transactions on Circ"its and )ystems for Video Technology5 Vol% :A5 $ss"e 85 pp%
B2@-B825 +"ly 1@@A%
7:@9 + ,cVeigh et% al%5 RA soft0are based real time ,!-G-1 video encoderS5 $---
Trans% C)VT5 Vol :@5 pp ::8<-::<35 ct% 1@@@%
7::9 %+% ,orris5 R,!-G-1: Nhere did it come from and 0hat is itUS5 $--
Collo'"i"m5 pp% :=:-:=B5 13 +an% :;;B%
7:19 !% V"n(elmann and #% Valva5 RCed"ced Complexity #%123 to ,!-G-1
TranscoderS5 $CC- 1@@85 pp% :-15 +an% 1@@8%
7:A9 /% Vamaci and H% Alt"nbasa.%5 R!erformance Comparison of the -merging
#%123 Video Coding )tandard 0ith the existing standardsS5 $C,-5 Vol%:5 pp% A3B-
A3<5 +"ly 1@@A%
7:39 +% Win5 C% Lin and ,% )"n 5 RDigital Video TranscodingS5 !roceedings of the
$---5 Vol% ;A5 $ss"e :5pp <3-;85 +an% 1@@B%
7:B9 A% Vetros5 C% Christopo"los and #% )"n5 RVideo transcoding architect"res
and techni'"es: an overvie0S5 $--- )ignal !rocessing ,aga(ine5 Vol% 1@5 $ss"e
15 pp :<-1;5 ,ar% 1@@A%
7:29 R,!-G-1S5 Ni.ipedia5 Deb% :35 1@@<%
Available at Xhttp:==en%0i.ipedia%org=0i.i=,pegM1Y
7:89 R$ntrod"ction to ,!-G 1 Video CompressionS
Available at Xhttp:==000%bretl%com=mpeghtml=codecdia:%#T,Y
7:<9 R#%123=,!-G-3 AVCS5 Ni.ipedia5 Deb% :<5 1@@<%
Available at X http:==en%0i.ipedia%org=0i.i=#%123Y
7:;9 R#%123 A ne0 Technology for Video CompressionS G Available at X
'ttp;@@999!nuntius!com@tec'no$ogy3!'tm$Y
71@9 C% !eriera5 R-fficient transcoding of ,!-G-1 to #%123S5 ,asterFs thesis5
38
Dec% 1@@B5 -- Department5 *niversity of Texas at Arlington%
71:9 R#%121 : $nformation technology - Generic coding of moving pict"res and
associated a"dio information: VideoS5 $nternational Telecomm"nication *nion5
1@@@-@1%
Available at X http:==000%it"%int=rec=T-C-C-#%121Y
7119 R,!-G-1 Nhite paperS5 !innacle Technical Doc"mentation5 Version @%B5
!innacle )ystems5 Deb% 1;5 1@@@%
71A9 ,% Ghanbari5 R)tandard Codecs : $mage Compression to Advanced Video
Coding5S #ert(5 *V: $--5 1@@A%
7139 #%123 soft0are >version :A%1? obtained from:
Xhttp:==iphome%hhi%de=s"ehring=tml=Y
71B9 ,!-G-1 soft0are >version :1? obtained from:
Xhttp:==000%mpeg%org=,!-G=video=mssg-free-mpeg-soft0are%htmlY
7129 Test streams >Doreman5 /e0s5 Carphone? obtained from:
Xhttp:==000-ee%"ta%ed"=dip=Co"rses=--BAB2=eeMBAB2%htmY
7189 $mplementation )t"dies Gro"p5 R,ain Ces"lts of the AVC Complexity
analysisS5 ,!-G doc"ment /3;235 $)=$-C +TC::=)C1;=NG::5 +"ly 1@@1%
71<9 A% +och et al%5 R!erformance comparison of video coding standards "sing
Lagarangian coder controlS5 $--- $nt% Conf% of $mage !rocessing5 Vol% 15 pp% $$-
B@: to $$-B@35 )ept% 1@@1%
71;9 $% )ylvester5 RTranscoding: The f"t"re of the video mar.et depends on itS5 $DC
-xec"tive Brief5 /ov% 1@@2%
Available at X 'ttp;@@999!ed.
c'ina!com@)R2-C,ES@200"1=5@2@200"1=5*0A)A)5CA1A*2!P/BY
7A@9 C% #offner5 R,!-G-3 Advanced Video Coding emergesS5
Available at X 'ttp;@@999!t%tec'no$ogy!com@features@2ec'.Corner@BAoffner.
03!09!05!s'tm$Y
7A:9 )% Nagston and A% )"sin5 R$! core for an #%123 Decoder )oCS5 1@@85
Available atX 999!us!design.reuse!com@ne9s@CidD*5+#"EprintDyesY
7A19 )% Vrishnamachari and V% Hang5 R,!-G-1 to #%123 Transcoding: Nhy and
#o0US5 Dec% :5 1@@25
Available at X
'ttp;@@&roadcastengineering!com@infrastructure@&roadcastingAmpegAtransc
odingA9'y@inde8*!'tm$Y
39