Graphics Reference
In-Depth Information
In video coding standards such as HEVC, the de-quantization process and inverse
transforms are specified, while the forward transforms and quantization process are
chosen by the implementer (subject to constraints on the bitstream).
This chapter is organized as follows. Section
6.2
describes the two transform
typesusedinHEVC:the
core
transform based on the discrete cosine transform
and the
alternate
transform based on the discrete sine transform. Design principles
used to develop the transform are also highlighted to provide insight into the
transform design process which considered both coding efficiency and complexity.
In Sect.
6.3
, the HEVC quantization process is described. Topics covered in this
section include the actual quantization and de-quantization steps, quantization
matrices, and quantization parameter derivation. Section
6.4
provides an overview
of the three special coding modes in HEVC (I_PCM mode, Lossless mode, and
Transform skip mode) that modify the transform and quantization process by either
skipping the transform or by skipping both transform and quantization. Sections
6.5
and
6.6
provide complexity analysis and coding performance results respectively.
HEVC Transform
1
6.2
The HEVC standard [
16
] specifies core transform matrices of size 4
4, 8
8,
16
16 and 32
32 to be used for two-dimensional transforms in the context
of block-based motion-compensated video compression. Multiple transform sizes
improve compression performance, but also increase the implementation complex-
ity. Hence a careful design of the core transforms is needed.
HEVC specifies two-dimensional core transforms that are finite precision
approximations to the inverse discrete cosine transform (IDCT) for all transform
sizes. Note that because of the approximations, the HEVC core transforms are not
the IDCT. The fact that an IDCT is not used does not necessarily make the HEVC
core transforms imperfect. In fact, the finite precision approximations are desirable
as explained in the next two paragraphs. The main purpose of the transform is
to de-correlate the input residual block. The optimal de-correlating transform is
the Karhunen-Loeve transform (KLT) [
22
] and not necessarily the DCT. This
is especially true for the coding of 4
4 luma intra-prediction residual blocks
where HEVC specifies an alternate 4
4 integer transform based on the discrete
sine transform (DST) [
24
]. Note that only the inverse transforms are specified in
the HEVC standard and the forward transforms are not. So an encoder may get
additional coding efficiency benefits by using the actual inverse rather than the
transpose of the inverse transform.
1
Portions of this section are © 2013 IEEE. Reprinted, with permission, from M. Budagavi,
A. Fuldseth, G. Bjøntegaard, V. Sze, M. Sadafale, “Core Transform Design in the High Efficiency
Video Coding (HEVC) Standard,”
IEEE Journal
of Selected Topics in Signal Processing,
December 2013.
