Early TU Split Termination in HEVC Based on Quasi-Zero-Block
Yunyu Shi, Zhiyong Gao, Xiaoyun Zhang
Institute of Image Communication and Network Engineering
& Shanghai Key Laboratory of Digital Media Processing and Transmissions
Shanghai Jiaotong University
Shanghai, China
yunyushi@sjtu.edu.cn
Abstract—The new introduced concept, transform tree or
residual quadtree (RQT) in HEVC standard, brings high
coding performance along with high computational complexity
because the optimal TU partition is estimated in a rate-
distortion sense at the encoder sides by testing all kinds of
partitions. This paper focus on early TU split termination in
terms of the tradeoff between computation and coding quality.
An early TU split termination scheme based on quasi-zero-
block (QZB) is proposed, which is defined by two aspects about
the quantized transform coefficients; the sum of all absolute
coefficients and the number of nonzero coefficients. Besides,
the selective probability of every TU depth is calculated and
analyzed. Experimental results show that the proposed method
(HM-QZB) could achieve 22.8% reduction in encoding time
and 50.59% reduction in TU processing time compared to the
HEVC test model HM10.0 encoder with about 0.04dB BD-
PSNR loss in coding performance.
Keywords-HEVC; early termination; TU split; residual
quadtree (RQT); quasi-zero-block (QZB)
I. INTRODUCTION
HEVC (High Efficiency Video Coding standard) project
has been conducted by JCT-VC (Joint Collaborative Team
on Video Coding) since 2010, working together in a
partnership between ITU-T and ISO/IEC. The main goal of
HEVC video coding standard is to provide a doubling in
compression efficiency with respect to existing standards,
such as H.264/AVC High Profile --- in the range of 50% bit
rate reduction for equal perceptual video quality [1].
Reference software, called the HEVC Test Model (HM), is
being developed along with the standard. At the time of
writing, the first edition of the HEVC standard has been
finalized in January 2013 [2] and the current version of HM
is HM10 [3].
Unlike H.264/AVC, a significant difference of HEVC is
the use of a more flexible quad tree structure based on a
coding tree unit (CTU), which has a size selected by the
encoder and can be larger than a traditional macro block.
Four different concepts, Coding Tree Unit (CTU), Coding
Unit (CU), Prediction Unit (PU) and Transform Unit (TU)
are introduced in HEVC. This variable-size, adaptive
approach is particularly suited to larger resolutions and plays
a major role in the substantial performance gains exhibited
by HEVC relative to previous video coding standards[4][5].
Meanwhile, it brings high computational complexity because
the optimal partitioning is estimated in a rate-distortion sense
at the encoder sides by testing all kinds of partitions.
To reduce the computational complexity of the encoding
process, most existing approaches reduce the number of CUs
or PUs to be tested [6]. References [7] is a typical example
related to fast HEVC encoding methods by proposing a fast
CU size decision algorithm based on motion homogeneity
checking, RD cost checking and SKIP mode checking to skip
motion estimation on unnecessary CU sizes. However, early
TU determination has rarely been exploited to reduce the
computational complexity in the HEVC encoder. Reference
[6] proposes an early TU decision method for fast video
encoding in HEVC by pruning the TUs at an early stage
based on the number of nonzero DCT coefficients. Two early
termination schemes for both TU Merge and TU Split
procedures are applied respectively for nonzero-blocks [7].
There are certain similarities among those previous methods
and they are tested with earlier HM version, such as HM3.0.
This paper focus on early TU split termination in terms of
the tradeoff between computational complexity and
compression quality. An early TU split termination based on
quasi-zero-block is proposed, where the concept of the quasi-
zero-block is defined by the number of nonzero quantized
transform coefficients and the sum of all absolute
coefficients. Besides, the selective probability of every TU
depth is calculated and analyzed. Experimental results show
that the proposed method achieves a 22.8% reduction in
encoding time and 50.59% reduction in TU processing time
compared to the HEVC test model HM10.0 encoder with
about 0.04dB BD-PSNR loss in coding performance.
The rest of the paper is organized as follows. Transform
Unit structure of HEVC is the topic of Section II. Section III
describes the proposed early TU termination method in detail.
The experimental results of the proposed method compared
with other prior similar method are presented and analyzed in
Section IV. Section V concludes the paper briefly.
II. RQT
STRUCTURE OF HEVC
In the main profile of HEVC, a slice is partitioned into
multiple CTU which are allowed to have size from 8×8 up to
64×64 [4]. A CTU includes one luma coding tree blocks
(CTB), two chroma CTBs, and syntax specifying coding data
and further subdivision. Each leaf node of the coding tree is
called CU with coding blocks (CB). The CU can have
multiple PU and a tree of TU. Similarly, each CB is split into
3rd International Conference on Electric and Electronics (EEIC 2013)
© 2013. The authors - Published by Atlantis Press