Image Processing Reference
In-Depth Information
efficient and increases the encoding performance of the complete proposed method
(including the adaptation of NDBV and IVMP to a full parallax structure, the two
IPMC, the DMV bi-prediction and the inter-view derivation of the second DV) up
to 2.9% and 4.2% for the sequence
Akko&Kayo
, respectively, with 3
3 and with
11
5 views.
Conclusion
Current 3D video technologies are not able to provide the viewer with a
smooth motion parallax, which is a very important cue in the perception of
depth. Integral imaging and Super Multi-View video are 3D technologies that
allow creating a light-field representation of a scene by capturing many
viewpoints of that scene. The large number and the density of the views
provide a smooth motion parallax in horizontal and potentially vertical
dimensions.
However the large amount of data and its specific structure represent a
challenge for future 3D video services and broadcasts. New efficient coding
technologies are required. In the case of full parallax 3D content, several
improvements are possible to exploit the vertical dimension. For integral
imaging, many methods have been proposed based on 3D-DCT and wavelet
transforms. Other methods propose to extract the views from an integral
image and to encode them with standard encoders (e.g., HEVC) and their
3D extensions (e.g., MV-HEVC and 3D-HEVC). Moreover, improvements of
these extensions have been proposed to encode content with horizontal and
vertical parallax.
In this chapter, an inter-view reference picture structure adapted to 3D
light-field video content with full motion parallax is proposed. Its main
features are the use of both horizontal and vertical inter-view references,
and the minimal distance between the coded and the reference views. The
proposed structure outperforms a basic anchor by up to 29.1% in BD-rate,
showing the impact of the use of both horizontal and vertical directions in the
inter-view reference picture scheme. Improvements of 3D-HEVC coding
tools NBDV and IVMP are also proposed, in order to exploit both horizontal
and vertical directions in a full parallax configuration, providing BD-rate
gains up to 4.2%. The results of the proposed methods show that exploiting
efficiently both horizontal and vertical dimensions of full parallax SMV
content at the coding structure and coding tools level can significantly
improve the compression performance.
Acknowledgements The Coast sequence is provided by Orange Labs. The Akko&Kayo
sequence is provided by Fujii Laboratory at Nagoya University.
