Adaptive Boundary Dependent Transform
Optimization for HEVC
Juanting Fan
1, 2
, Jicheng An
3
, Shanshe Wang
1, 2
, Nan Zhang
4
, Ruiqin Xiong
1, 2
, Siwei Ma
1, 2
, Shawmin Lei
3
1
Institute of Digital Media, Peking University, Beijing, China
2
Cooperative Medianet Innovation Center, Shanghai, China
3
MediaTek Inc., Beijing, China
4
School of Biomedical Engineering, Capital Medical University, Beijing, China
{jtfan, sswang, rqxiong, swma}@jdl.ac.cn, {jicheng.an, shawmin.lei}@mediatek.com, zhangnan@ccmu.edu.cn
Abstract—High Efficiency Video Coding (HEVC) adopts hybrid
transform coding scheme to improve the transform performance.
However, it does not consider the influence of the prediction unit
boundary. This paper proposes an adaptive boundary dependent
transform optimization scheme for HEVC. Based on the
transform unit boundary type identification by checking whether
it lies at a prediction unit boundary or not, some additional
transform kernels, including DCT-IV, are incorporated for
transform. Moreover, the additional adopted transform kernels
are generated by re-using the existing kernels in HEVC.
Therefore, the butterfly structure for transform implementation
can be preserved to facilitate parallel computing, and
computational complexity increasing can be avoided.
Experimental results show that the coding performance
improvements can be up to 0.96% and 0.64% for low delay P and
random access testing configurations respectively.
I. INTRODUCTION
Transform coding is widely adopted in image and video
compression. Its main aim is to reduce the inherent spatial
redundancy between adjacent pixels. It has been proved that the
Karhunen-Loeve Transform (KLT) is the optimal transform
coding method for video and image compression that can
reduce the correlation [1]. However, practical use of KLT is
limited due to its high computational complexity. Discrete
Cosine Transform (DCT) and Discrete Sine Transform (DST)
have been testified to offer good energy compaction for image
compression and can achieve coding performance verge on
KLT [2]. However, DCT and DST consist of a bunch of
different transform kernels, it is always a popular research topic
to determine the optimal transform kernel [3].
In the past researches, it has been studied that a very few
typical transform kernels could be practically adopted in image
and video coding. The DCT-II is one of these. It was first
adopted into H.264/AVC, bounded with a hybrid transform
coding scheme with typical block sizes from 4×4 to 16×16
depending on the size of prediction blocks after intra/inter
prediction [4] [5]. And it was definitely used for High
Efficiency Video Coding (HEVC) as well, with an extensional
range of block size, including 4×4, 8×8, 16×16 and 32×32 [6]-
[8].
In recent researches, many research efforts have been
devoted to optimizing the transform kernel determination to
fully exploit signal correlation to achieve performance
improvement. In [9] and [10], based on the distribution of
intra-predicted residuals which is denoted as boundary
information, an optimization scheme for transform kernel
determination is proposed. It is stated that DST-VII is more
effective for the transform unit at the boundary. DST-VII was
finally adopted to HEVC [8]. However, due to high complexity
of DST-VII, it was only applied for 4x4 Transform Units (TU).
Thus, the overall HEVC intra-predicted hybrid transform
coding scheme switches between DST-VII and the
conventional DCT-II for 4x4 transform units. For other
transform units, only DCT-II is employed. In [11], Mode
Dependent Directional Transform (MDDT) is fully
investigated and it can be seen that MDDT can bring much
performance improvement with high computational complexity.
In [12], Zhao etc. propose a Rate Distortion Optimization
Transform (RDOT) by providing more transform kernels.
Better performance improvement can be achieved with much
higher computational complexity.
In order to avoid high computational complexity, only 4-
point (4×4) fast DST-VII is adopted into HEVC [8], and Cohen
etc. propose a simple MDDT scheme for intra-predicted
transform with little performance loss [13]. In [14] [15], fast
algorithms for DCT and IDCT (Inverse Discrete Cosine
Transform) are proposed for butterfly structured
implementation considering parallel computing and hardware
design.
So, the adoption of multiple types of transform kernels
improves compression performance significantly. However,
computational complexity increases much. In this paper, we
propose a Boundary Dependent Transform (BDT) optimization
scheme. For the given inter prediction unit, different DST/DCT
kernels are adaptively determined for transform based on the
boundary information to improve the coding performance
without any computational complexity increasing.
The rest of this paper is organized as follows. Section II
gives a detailed description of inter-predicted residual analysis.
Then, the proposed adaptive boundary dependent transform
scheme is provided in section III. Section IV shows our
experimental results. Finally, we conclude this paper in section
V.
II. INTER-PREDICTED RESIDUAL ANALYSIS
Inter prediction plays a key role in video coding. It can
improve coding performance significantly. However, for a