Improved Methods for Initializing R-Q Model Parameters and Quantization
Parameter in H.264 Rate Control
Xiaodong Liu, Qionghai Dai, Chonggang Fu
Broadband Network & Multimedia Research Center
Graduate School at Shenzhen, Tsinghua University, 518055, Shenzhen, P. R. China
Abstract
Rate control plays an important role in video
coding and transmission. Current initialization of
Quantization Parameter (QP) in H.264 rate control
takes the same value for images with different
complexity as long as the bit number per pixel is the
same which is different from the real-world situation.
In this paper, we propose a new R-Q model and QP
initializing method to improve the precision of rate
control. In this method, the quantization parameter
and R-Q model parameters are calculated from the
first two frames of video sequence. Some formulas are
presented to implement the process in this paper.
Experimental results show that the output bit rate
achieved by our proposed one is more close to the
target bit rate while the reconstructed quality is
comparable with low computing complexity.
1. Introduction
Among real time video applications such as video
surveillance and video conferencing, efficient video
compression plays an important role for the video
communication. H.264/AVC is the latest international
video coding standard developed by the Joint Video
Team (JVT) and the novel features of H.264/AVC
have significantly improved the coding efficiency
when compared to previous video coding standards[1].
In addition to coding efficiency, Rate Control (RC) is
also an important issue in video coding, especially for
real time communication. Rate control aims to achieve
good perceptual quality given the transmission
bandwidth constraint[2]. Usually, RC regulates the
coded bit stream by adjusting Quantization
Parameter(QP). To achieve this, a rate-quantization (R-
Q) model is used to represent the coded bits by means
of QP and the Mean Absolute Difference (MAD) of a
residual frame. Unfortunately, using MAD for R-Q
modeling causes the chicken-and-egg dilemma to the
H.264 standard [3].
Several RC algorithms have been developed to
overcome the H.264 RC chicken-and-egg dilemma.
One rate control scheme with temporal MAD
prediction method and R-Q quadratic model was
proposed in JVT-G012 [3] as follows:
[] [ ]
21,
1 aiMADaiMAD
actualpredlinear
+−×=
(1)
[] []
[]
[]
[]
[]
⎟
⎟
⎠
⎞
⎜
⎜
⎝
⎛
+×=
2
21
iQ
iX
iQ
iX
iMADiR
texture
(2)
where
[]
iMAD
predlinear,
and
]1[ −iMAD
actual
are
the predicted MAD for the current frame i and the
actual MAD of previous frame i-1, respectively. a
1
and
a
2
are two parameters that will be updated after coding
each frame.
[]
iR
texture
is the bit rate for the texture
information of the current ith coding frame.
[]
iMAD
and
[]
iQ
are its coding complexity measure and
quantization step size, respectively.
[]
iX
1
and
[]
iX
2
are the first-order and second-order parameters of this
quadratic R-Q model, which would be updated after
encoding every frame.
Another RC algorithm is proposed for H.264 low
delay video communications in [4], where an improved
MAD prediction model and a linear R-Q model are
applied as follows:
[] [] [] []
iZiMADiZiMAD
roughspatpred 21,
+×= (3)
[] []
[]
[]
[]
iX
iQ
iX
iMADiR
p
adaptpredsum 2
1
,
+×= (4)
where
[]
iMAD
rough
is a measure for evaluating the
difference between the current original frame and the
previous reconstructed frame.
spatpred
MAD
,
is the
spatial predicted MAD of the current frame.
[]
iZ
1
and
[]
iZ
2
are the first-order and zeroth-order parameters
of this linear prediction model, which would be
2009 World Congress on Computer Science and Information Engineering
978-0-7695-3507-4/08 $25.00 © 2008 IEEE
DOI 10.1109/CSIE.2009.243
320
2009 World Congress on Computer Science and Information Engineering
978-0-7695-3507-4/08 $25.00 © 2008 IEEE
DOI 10.1109/CSIE.2009.243
320