DISPARITY COMPENSATION BASED 3D HOLOSCOPIC IMAGE CODING USING HEVC
Deyang Liu, Ping An, Ran Ma, Liquan Shen
School of Communication and Information Engineering, Shanghai University, Shanghai 200072, China
Key Laboratory of Advanced Displays and System Application, Ministry of Education, Shanghai, China
ABSTRACT
3D holoscopic image can provide true 3D content by
reproducing the light rays of the 3D scene and is regarded
as a promising technique for future 3D TV. Disparity
information is paramount intrinsic characteristic of the 3D
holoscopic image. In this paper, a disparity compensation
based 3D holoscopic image coding algorithm using HEVC
is put forward. In order to drive an authentic disparity, we
expand the available area outwards in the disparity
matching process instead of dividing the current coding
block into four parts. Experimental results demonstrate that
the proposed method can obtain considerable gains over
original HEVC intra-prediction and the classical TMP
algorithm with acceptable complexity increasing compared
to original HEVC standard.
Index Terms— 3D holoscopic image, disparity
compensation, image prediction, image coding, HEVC
1. INTRODUCTION
Holoscopic imaging, also referred to as integral imaging,
going back to the pioneering work of Lippmann, can
provide true 3D content by reproducing the light rays of the
3D scene, for which it can minimize some uncomfortable
feelings such as eye strain or headache when people focus
on the screen for a long time. In the simplest form, the
holoscopic imaging system with full parallax consists of a
lens array mated to a digital sensor. Each lens of the lens
arrays can capture perspective views of the 3D scene. For
this reason, 3D holoscopic imaging offers 3D feelings for
more than one person, independent of viewer’s positions
and it is regarded as a promising technique for future 3D
TV.
In order to supply immersive experience for multiple
viewers, much higher image resolution is needed to fit high
definition requirements. Consequently, effective coding
tools are desirable for such particular type of content.
High Efficient Video Coding (HEVC) [1], a new
standard for video coding, developed by Joint Collaborative
Team on Video Coding (JCT-VC), has adopted many
advanced encoding tools and can significantly improve the
compression performance of high definition videos with
half of the bit rate saved compared to the H.264/Advanced
Video Coding (AVC) [2] for the same perceptual video
quality. Therefore, HEVC standard promises to improve the
coding efficiency for holoscopic imaging coding.
Several 3D holoscopic image coding schemes have
been proposed in the literatures. In [3-4], a prediction
coding for 3D holoscopic content, combining the flexible
coding tools from HEVC with self-similarity estimation
concept [5] which exploits the special arrangement of 3D
holoscopic image, is proposed. In this scheme, new
prediction modes are added to explore the particular
structure of 3D holoscopic content. This method can
achieve a good performance, but has to modify the bit
stream structure.
In [6] a locally linear embedding (LLE)-based
prediction algorithm is introduced into the HEVC encoder
to handle 3D holoscopic image. In this paper, some
directional intra prediction modes of the HEVC are
replaced by a more efficient predictive framework based on
LLE technology. The main idea of the LLE [7] based
prediction is to estimate the coding block by using a linear
combination of k-nearest neighbors (k-NN) patches,
determined in the causal coded and reconstructed region of
the image.
The algorithms that are mentioned above can be
interpreted as the improved algorithm based on two
classical methods: intra displacement compensation (IDC)
and template matching prediction (TMP). The two
approaches all try to exploit the self-similarities within a
picture. IDC [8] reuses the reconstructed or decoded block
patches to drive a best intra displacement vector, and then
the best matching block patch is used to predict the current
block. The best displacement vector per block must be sent
to the decoder as overhead. TMP [9] obtains the best
candidate prediction block through measuring the similarity
between the surrounding neighboring pixels and the
candidate template patches which are already reconstructed
or decoded within a searching range. Typically the Sum of
Absolute Differences (SAD) or Sum Squared Errors (SSE)
is chosen as the criterion to retrieve those matching blocks
with higher similarity. After finding the best candidate
template patch, the prediction block, therefore, is used to
predict the current coding block. Different from IDC, TMP
201978-1-4799-1948-2/15/$31.00 ©2015 IEEE ChinaSIP 2015
资源评论
