Digital Signal Processing 44 (2015) 37–46
Contents lists available at ScienceDirect
Digital Signal Processing
www.elsevier.com/locate/dsp
Fast mode decision algorithm for 3D-HEVC encoding optimization
based on depth information
Qiuwen Zhang
∗
, Xiao Wang, Xinpeng Huang, Rijian Su, Yong Gan
College of Computer and Communication Engineering, Zhengzhou University of Light Industry, Zhengzhou 450002, China
a r t i c l e i n f o a b s t r a c t
Article history:
Available
online 16 June 2015
Keywords:
3D-HEVC
Mode
decision
Depth
map
Correlation
The advanced 3D extension of the High Efficiency Video Coding standard (3D-HEVC) encodes 3D videos
in the multi-view video plus depth format (MVD) and introduces new prediction modes to improve the
coding performance of dependent views and depth data. Similar to HEVC, the mode decision process in
3D-HEVC is performed using a large number of prediction modes to find the one mode with the least
rate distortion (RD) cost using Lagrange multiplier. These techniques achieve the highest possible coding
efficiency, but also bring high computational complexity, which may limit the use of 3D-HEVC in practical
applications. In this paper, we propose a fast mode decision algorithm based on the depth information to
reduce 3D-HEVC computational complexity. Since the video texture and depth map are highly content-
dependent,
it is not efficient to use all the prediction modes in 3D-HEVC mode decision procedure.
There exist high correlations of the coding information when two CUs have the similar depth values.
Based on the correlations, we propose two efficient mode decision techniques, 1) an early coding level
determination is developed based on the spatially neighboring CUs depth values; 2) an adaptive mode
size decision is introduced based on the correlations between the motion characteristics and the depth
values. Experimental results show that the proposed algorithm can save 74% computational complexity
on average with only 1.47% increase on bit rate or 0.06 dB loss on peak signal-to-noise ratio (PSNR)
compared with the original 3D-HEVC encoder.
© 2015 Elsevier Inc. All rights reserved.
1. Introduction
3D video coding (3DVC) has attracted great attention from in-
dustries
and research institutes [1–4]. 3D extension of high effi-
ciency
video coding (3D-HEVC) [5] is used to encode multi-view
video plus depth map format (MVD) [6] streams for 3D playout
systems such as 3D television (3DTV) [7], free viewpoint television
(FTV) [8], and 3D digital cinema applications. 3D-HEVC is an ex-
tended
version of high efficiency video coding (HEVC) [9] that im-
proves
the performance of MVD. Different from single-view video
coding, inter-view and inter-component redundancies exist in mul-
tiview
video coding [10]. These newly added prediction modes
achieve the highest possible coding efficiency, but with signifi-
cant
computational complexity increase which obstruct 3D-HEVC
from practical applications. Therefore, speed-up algorithms, which
can reduce the complexity of mode decision of 3D-HEVC encoding
*
Corresponding author at: College of Computer and Communication Engineering,
Zhengzhou University of Light Industry, No. 5 Dongfeng Road, Zhengzhou 450002,
China. Fax: +86 037186609561.
E-mail
address: zhangqwen@126.com (Q. Zhang).
without compromising coding efficiency, are essential for realizing
related real-time applications.
Recently,
a number of fast mode decision algorithms [11–18]
have
been proposed to reduce the computational complexity for
the previous video coding standard such as H.264/AVC and its
extension of multi-view video coding (MVC). Afast algorithm to
jointly optimize mode decision and motion estimation (ME) is pro-
posed
in [11] to reduce complexity of H.264 mode decision. Afast
mode decision based on motion spatial property is presented in
[12] to decide best prediction mode in H.264 inter coding. A hi-
erarchical
fast algorithm is proposed in [13] to speed up the in-
tra
mode decision process by reducing the number of H.264/AVC
modes for each macroblock (MB). Based on the motion dominance
and
relative positioning, a fast inter mode decision algorithm is
introduced in [14] where motion estimation and compensation
modes are directly selected using phase correlation information in-
stead
of exhaustive mode selection with all possible large modes.
The methods presented in [15] and [16] use the motion homo-
geneity
or coding mode complexity to reduce encoding complexity
of MVC. A mode correlation-based approach in [17] is applied to
the mode decision procedure in MVC encoders to speed-up the en-
coding
computation. In our previous work [18], the texture video
http://dx.doi.org/10.1016/j.dsp.2015.06.005
1051-2004/
© 2015 Elsevier Inc. All rights reserved.