A Fast CU Decision Algorithm for MV-HEVC
Peicheng Wang, Xingang Liu, and Binfei Shao
School of Electronic Engineering
University of Electronic Science and Technology of China
Chengdu, China
pcwanguestc@163.com
Abstract—MV-HEVC is a 3D extension of the High Efficiency
Video Coding (HEVC) standard, which is developed to improve
the coding efficiency of multi-view video. However, this improved
efficiency leads to greater computational complexity. Recently,
how to reduce the encoding time for multi-view coding is
becoming a hot issue in real-time applications. In this paper, a
fast coding unit (CU) decision algorithm is proposed by utilizing
spatiotemporal and inter-view information, including the
maximum depth level of related CUs and motion vectors. Firstly,
the maximum depth level of current CU is predicted by using
spatiotemporal and inter-view CUs. Then, it will be adjusted by
checking whether the current CU has the same motion feature
with nearby spatial CUs. Experimental results show that our
proposed algorithm can achieve up to 67.16 % time reduction
with a little degradation in encoding efficiency.
Keywords—3D video coding; Early CU decision; MV-HEVC;
Spatiotemporal and inter-view correlation.
I. INTRODUCTION
Multi-view video and 3D video (3DV) provide more
immersive viewing quality of experience compared to
traditional 2D video. In the near future, any display devices
will be developed to support stereoscopic video. Although
three-dimensional video can be realized with current popular
video coding standard, H.264/AVC [1], to improve multi-view
video coding efficiency and provide other novel functionalities,
new multi-view and 3DV standards [2] have been proposed.
The main functionality of the new 3DV standards is able to
generate arbitrary virtual views according to the input views.
The most popular multi-view video coding standard (MVC),
which is an extension of H.264/AVC, was designed to encode
stereoscopic video with two or more views. In MVC, the inter-
view correlation is exploited to improve multi-view video
coding efficiency by using pictures in a different view as
reference pictures of the current view. But no modifications at
the macroblock have been changed. Only high-level syntax,
such as sytax, semantics and reference pictures in a level than
block level need to be revised.
Recently, a new video compression standard, known as the
High Efficiency Video Coding (HEVC) [3], is introduced by
Joint Collaborative Team. HEVC has been designed to cope
with the increased video resolution and increased use of
parallel processing architectures. Compared with the coding
efficiency of H.264/AVC, it enables approximately 50% bit-
rate reduction for equal perceptual video quality due to
utilizing many advanced tools. Among these, one of the most
powerful tools to improve coding efficiency of HEVC is the
hierarchical coding structure, which is based on the quad-tree
structure of coding unit (CU). Also, each CU block has the
prediction unit (PU) blocks of symmetric and asymmetric
sizes and the transform unit (TU) blocks of quad-tree
partitions. Fig. 1 shows an example of CU, PU, and TU split
structure in HEVC.
Fig. 1. Example of CU, PU, and TU split structure.
As shown in Fig. 1, the largest CU is called coding tree unit
(CTU). During the motion estimation (ME) of encoding
process, each CTU block can be split into smaller CU blocks
from 64x64 to 8x8 in rate distortion optimization process. The
CU size can vary from the CTU size to the smallest CTU
(SCU) to get the best appropriate CU. PU is a basic unit for
ME where all modes, such as Skip, Merge, 2Nx2N, 2NxN,
Nx2N, NxN,2NxnU, 2NxnD, nRx2N and nLx2N, must be
checked to find the best PU for the current CU. In addition,
NxN mode only is checked when the current CU is the SCU.
Finally, TU in 2Nx2N CU block can be split in a quad-tree
structure and each split TU size can further be split recursively.
After the new video compression standard was released, the
development of multi-view extension of HEVC, namely, MV-
HEVC [4], had been started in Joint Collaborative Team on
3DVCoding Extension Development (JCT-3V). Similar to
MVC, MV-HEVC is a high-level syntax only extension of
HEVC. In MV-HEVC, inter-view prediction is enabled in a
way that no block-level changes are required. Fig. 2 shows a
prediction structure of MV-HEVC. View 0 is a base view, the
reference pictures of which are from its coded pictures. But
2015 IEEE International Conference on Smart City/SocialCom/SustainCom together with DataCom 2015 and SC2 2015
978-1-5090-1893-2/15 $31.00 © 2015 IEEE
DOI 10.1109/SmartCity.2015.74
217