3D Video Representation and Coding - Novel 3D Media Technologies - page 41

All the techniques used to provide a faster encoder result in some loss in coding
gain. This occurs because of the predictive nature of the used algorithms that are
sub-optimal in nature. The losses are well contained and the gain in terms of speed
is more beneficial than the small loss in quality, which is typically difficult for the
viewer to distinguish.
Improved Coding Efficiency
The 3D video coding extension (3DVC) of the high efficiency video coding that is
currently being developed by MPEG takes into consideration the similarities
between the texture and depth MVVs during encoding. The base view can be
independently decoded with a HEVC codec for compatibility purposes but the
other views, including depth MVVs, depend on this stream. Additional coding
tools are therefore applied together with an inter-component prediction scheme
that uses the already coded components at the same time instant [ 24 ]. These
modifications include motion parameter inheritance where the depth map coding
inherits the partitioning of the blocks and the motion parameters from the texture
image. These motion vectors are quantized to full-sample precision compared to the
quarter-sample accuracy in the texture video. The decision on whether to inherit
the vectors or not is done for each block. The rate distortion optimization takes into
consideration the synthesized view, given that the main need for multiview video-
plus-depth coding is to generate virtual views for autostereoscopic displays and free
viewpoint applications. For a more detailed understanding of the multiview video
and depth data coding for 3D HEVC, the reader is referred to [ 24 ].
In order to save bandwidth, the depth video can be down-sampled at the encoder
and then up-sampled again at the decoder. A joint texture/depth edge-based
up-sampling algorithm is proposed in [ 52 ] to improve the quality of the edges in
the depth video and hence provide a gain in quality of the synthesized views. The
joint algorithm exploits the similar geometry and intensity of the depth data and the
corresponding texture image. Each sample is further adapted using weights to the
characteristics of the depth map to further improve the quality. Results reported in
[ 52 ] on H.264 show an improved coding efficiency.
Other work was directed to use Scalable Video Coding (SVC) to aid multiview
video-plus-depth coding. In [46, 53] the authors present an inter-view prediction
scheme which is similar to MVC together with an inter-layer motion prediction
scheme adopted from SVC. The latter exploits the correlation that exists between
object movements in both texture and depth video streams, thus the base layer is
used to carry the texture video while the coarse granularity scalability enhancement
layer carries the depth data. An improvement of 10-20 % in coding efficiency is
reported in [ 53 ] and up to 0.97 dB gain is reported in [ 54 ].