Game Development Reference
In-Depth Information
Chapter 5
Transform and Quantization
This chapter gives an overview of transform coding techniques, including four
sections. The first section introduces the basic principles and recent developments of
transform and quantization coding. In the second and third section, we provide more
details on the technical design of transform and quantization in AVS1 and AVS2.
The last section summarizes this chapter.
5.1 Related Works
5.1.1 Transform Coding
Transform coding, which can reduce the spatial correlation among residual signals
efficiently, has been widely applied in various video coding standards. Although
both prediction and transform target to remove the statistical redundancy existing in
the spatial/temporal domain, it is noted that prediction only handles the correlation
between the samples of the reference block and the samples of the current block,
while transform is able to further exploit the correlation of the samples within the
current block.
The one-dimensional transform process can be formulated as the following linear
operation:
y
=
x
·
T
,
(5.1)
where
x
and
y
represent the input and output data which are both vector of size
N
,
T
denotes the transform matrix whose size is
N
N
. To extend the transform to the
two-dimensional (2-D) sources, the following formulation is applied:
×
Y
=
C
·
X
·
R
,
(5.2)
