
Proceedings
of
ISCIT2005
Fast
Mode
Decision
for
Inter
Frames
in
JJ.
2
64/AVC'
Lidong
Xut
Xinggang
Lin
Department
of
Electronic
Engineering,
Tsinghua
University,
Beijing,
100084,
China
Tel:
+86-10-62781291,
Fax:
+86-10-62770317
E-mail:
remann97
4mails.tsinghua.edu.cn
Abstract-The
H.264/AVC
video
coding
standard
can
achieve
considerably
higher
coding
efficiency
than
any
previous
standards.
Unfortunately
this
comes
from
dramatically
high
coding
complexity
introduced
mainly
by
multi-mode
checking
scheme
when
coding
macroblocks.
Base
on
statistically
analyzing
the
best
mode
situations,
a
progressively
adaptive
fast
mode
decision
algorithm
is
proposed
in
this
paper
to
reduce
the
coding
complexity
of
inter
frames
in
H.264/AVC.
Experimental
results
show
that
the
proposed
algorithm
can
averagely
save
about
75.3%
coding
time
compared
with
the
reference
software
with
only
about
0.15%
bitrate
increase
and
0.08dB
PSNR
loss.
Keywords:
H.264/AVC,fast
mode
decision,checking
priority
I.
INTRODUCTION
The
emerging
H.264/AVC
video
coding
standard,
jointly
developed
by
ITU-T
VCEG
and
ISO/IEC
MPEG
for
various
applications
such
as
television
broadcasting,
digital
storage
media,
communication
and
video
conferencing
[1],
achieves
a
significant
coding
efficiency
gain
relative
to
other
existing
coding
standards,
e.g.
H.263,
MPEG-2
and
MPEG-4.
Under
the
same
distortion,
H.264/AVC
can
save about
50%
bitrate
as
compared
with
H.263
and
MPEG-4
[2].
The
high
coding
efficiency
of
H.264/AVC
comes
from
high
computationally
complex
features
among
which
the
enhanced
intra
and
inter
prediction
techniques
are
key
factors
to
the
success
of
H.264/AVC
[3].
In
H.264/AVC,
seven
block
types,
which
are
16x16,
16x8,
8x16,
8x8,
8x4,
4x8
and
4x4,
are
supported
for
coding
a
macroblock
(MB).
Corresponding
to
these
block
types,
there
are
five
inter
coding
modes,
which
are
SKIP/DIRECT
(MO),
inter_16x16
(Ml),
interl6x8
(M2),
inter_8x16
(M3)
and
P8x8
(M4),
and
two
intra
coding
modes,
which
are
intra_4x4
(M5)
and
intra
l6xl6
(M6).
The
P8x8
mode
includes
four
kinds
of
block
types:
8x8
(inter
8x8),
8x4
(inter
8x4),
4x8
(inter_4x8)
and
4x4
(inter_4x4).
For
intra
coded
frames,
only
intra_4x4
and
intra_16xl6
modes
are
available,
and
for
inter
coded
frames,
all
the
seven
coding
modes
are
available.
The
multi-mode-checking
scheme
introduces
very
high
coding
complexity
for
encoding
an
MB
and
brings
great
difficulty
to
real
time
applications.
So
many
fast
mode
decision
algorithms
are
proposed
to
accelerate
the
encoder.
Kim
[3]
and
Pan
[4]
proposed
some
schemes
for
fast
intra
mode
decisions.
But
in
video
coding,
most
of
the
frames
and
MBs
are
inter-coded.
So
fast
inter
mode
decision
can
bring
more
benefits
to
encoder
speedup.
Zhu
[5]
reduced
the
candidate
inter
modes
by
pre-encoding
a
down-
sampled
small
image.
This
method
can
reduce
about
50%
encoding
time
with
PSNR
decrease
less
than
0.2dB.
Lee
[6]
proposed
"cearly
SKIP
mode
decision"
and
"selective
intra
mode
decision"
techniques
which
reduced
about
30%
encoding
time
with
averagely
about
0.7%
bitrate
increase
and
0.03dB
PSNR
drop.
In
[7],
the
algorithm
proposed
by
Chen
can
save
about
67%
encoding
time
with
averagely
about
2.1%
bitrate
increase
and
O.1dB
PSNR
loss.
In
this
paper,
based
on
statistically
analyzing
the
best
mode
situations,
we
proposed
a
threshold
based
progressively
adaptive
mode
checking
algorithm
for
inter
frames,
i.e.
P
and
B
frames,
which
can
significantly
reduce
the
encoding
time
with
only
slight
coding
efficiency
loss
as
compared
with
the
reference
software.
The
rest
of
this
paper
is
organized
as
follows.
Section
II
statistically
analyzes
mode
decision
results
for
inter
frames.
Section
III
depicts
the
proposed
fast
mode
decision
algorithm.
The
experimental
results
are
shown
in
section
IV.
Finally
section
V
concludes
this
paper.
II.
MODE
DECISION
ANALYSIS
In
H.264/AVC,
the
mode
decision
is
performed
based
on
rate-distortion
optimization
(RDO)
which
is
done
by
minimizing
the
Lagrangian
function
J:
J(s,c,MODEIQP,4ODE)
=
SSD(s,c,MODEIQP)
(1)
+tODER(s,c,MODEIQP)
where
QP
is
the
quantization
parameter,
kMODE
is
the
Lagrange
multiplier
for
mode
decision,
SSD
is
the
sum
of
the
squared
differences
between
the
original
block
s
and
its
I
This
project
is
partially
supported
by
National
Natural
Science
Foundation
of
China
under
Grant
No.
60472028,
Doctoral
Program
Research
Foundation
of
Ministry
of
Education
of
China
under
Grant
No.
20040003015
and
LG
Electronics
Inc.
(Korea).
0-7803-9538-7/05/$20.00©2005
IEEE
418
Authorized licensed use limited to: Dalian University of Technology. Downloaded on January 2, 2010 at 03:25 from IEEE Xplore. Restrictions apply.