Fast coding tree unit decision for HEVC intra
coding
Caixia Bai
Computer science constitute of Tsinghua University
518055 Shenzhen, China
caixiabai@gmail.com
Chun Yuan
Computer science constitute of Tsinghua University
518055 Shenzhen, China
yuanc@sz.tsinghua.edu.cn
Abstract--- The Quad-tree structure in HEVC contributes
to significant coding efficiency improvement, especially for
high resolution videos. However, its recursive representation
increases the coding complexity as well. This paper brings
forward a fast coding tree uint algorithm to speed up the
original intra coding in HEVC. By analyzing the texture of the
CU, to filter out some unnecessary CU candidates. In the
paper, two variables are used to evaluate the texture of the
CU, that is the sobel gradient and MAD value. Combined these
two values, encoding of some CUs can be skipped.
Experimental results show that the algorithm obtains around
30% of time reduction with nearly unchanged PSNR and
bitrate stays almost the same compared with the original
HEVC intra coding.
KeyWordss—HEVC; intra coding; quad-tree structure
I. INTRODUCTION
The demands on high quality video service such as UHD
and 3D-UH are increasing, which requires acceleration in the
development of a video coding with high compression
efficiency. Since the April of 2010, a new intenational
standard names High Eifficiency Video Coding (HEVC) [1]
[2] is under standardization by JCT-VC. HEVC [3] aims to
improve coding efficiency which has a bit rate reduction of
50% at the same subjective image quality compared to the
H.264/MPEG-4 AVC High profile. A lot of new efficient
coding tools are adopted in HEVC, such as quad-tree
structure, extended intra prediction modes, advanced motion
vector prediction, adaptive loop filter and so on. HEVC [4]
[5]exploits three types of units, coding unit (CU), prediction
unit (PU) and transform unit (TU). CU is a basic coding unit
which can be split from the largest CU (LCU) to the smallest
CU (SCU). Each LCU can be split into four CUs recursively
up to the maximum allowable hierarchical depth. As shown
in Fig. 1. PU specifies the basic unit used for carrying the
information related to prediction process, including motion
estimation and rate-distortion optimization. TU is a basic unit
used for residual transform and quantization. After a
properly-sized PU is determined through testing various
coding modes, an appropriate size of TU is determined.
A. The quad-tree structure of HEVC
For HEVC, among the three units, CU is the most critical in
relation to the compression efficiency because it determines
the initial coding block size, which affects the performance
of the remaining processing units. In the HM encoder, each
frame begins its encoding from LCU. To select the optimal
partition mode for LCU, comparison of the rate-distortion
cost of all possible partition is needed. Flexibility of CU size
can efficiently adapt to the diversity of picture content to
bring coding gain.
Fig. 1 A partition of LCU and its corresponding coding quad-tree
On the one hand, setting LCU size to 64x64 can
outperform the size of 16x16 in nearly 12% bitrates
reduction on average. On the other hand, it imposes great
computational complexity on encoder to choose the optimal
CU size. Because of the maximum depth can be as deep as
four, in order to select the optimal CU size, it would cause
exponential growth in the number of CU encoding. Up to 85
(1+4+16+64) CU encoding are computed in order to find the
optimal CU partition with the smallest RDCost. Since each
CU is followed by a corresponding PU and TU computation,
the growing complexity poses a serious problem to the
encoding time.
B. Related work
Choi et al [6] raised a coding tree pruning based CU early
termination method to reduce the encoding time. If the