3D Holoscopic Video Representation and Coding Technology - Novel 3D Media Technologies - page 91

chosen for each test image: in the first set, patches of size 4 10,

respectively, for Plane and Toy and Laura images were chosen so as to have the
main object of the scene in focus (i.e., the object in the center of the image); in
addition to this, the second set with sizes of, respectively, 10

4, and 10

20 were
also considered to evaluate the influence of patch size parameter in the performance
of the proposed scalable coding scheme. For the sake of simplicity, the nine views
of each test image were coded independently with the HEVC with “Intra, main”
configuration [ 29 ].
Afterwards, each set of nine coded and reconstructed 2D views is processed to
generate a corresponding IL reference, which is ready to be included in the
reference picture buffer of the HEVC together with the SS reference.
Four different scenarios are tested and compared to evaluate the performance of
the described scalable 3D holoscopic coding:

10, 20

1. HEVC Simulcast : In this scenario, each 3D holoscopic test image is coded with
the HEVC using the “Intra, main” configuration [ 29 ].
2. 3DHolo Simulcast : Each 3D holoscopic test image is coded independently using
the self-similarity compensation prediction in the HEVC, as proposed in [ 25 ];
3. Scalable : Each 3D holoscopic test image is coded with the scalable coding
solution presented in [ 15 ], considering the two different view rendering algo-
rithms (Basic Rendering and Weighted Blending) with the different patch sizes;
4. 3DHolo Scalable : In this scenario, each 3D holoscopic test image is coded with
the proposed combined scalable coding solution, considering the two different
view rendering algorithms (Basic Rendering and Weighted Blending) with the
different patch sizes.

The results are presented in Tables 5.1 and 5.2 in terms of Bjontegaard Delta
(BD) measurement method [ 30 ] of PSNR and rate (BR) for each test image.
Additionally, the RD performance for Plane and Toy is also presented in
Fig. 5.16 to better illustrate the gains presented in Tables 5.1 and 5.2 . Since only
the performance of the Second Enhancement Layer is analyzed (as justified in
Sect. 5.4 ), the results for each view rendering algorithm, as well as for each used
patch size, are properly presented in different charts as it refers to different content
in the first two hierarchical layers of the scalable coding architecture. Moreover, it
should be noticed that the coding efficiency is analyzed here in terms of the gain

Table 5.1 BD-PSNR and BD-BR test results for Plane and Toy
3D Holo Scalable (PSNR [dB] and BR [%])
Basic
Rendering,
patch size 4

Weighted
Blend, patch
size 4

Basic
Rendering,
patch size 10

Weighted
Blend, patch
size 10

PSNR BR PSNR BR PSNR BR PSNR BR

HEVC Simulcast 1.48 20.13 1.58 21.16 1.68 21.58 1.68 21.58

3DHolo Simulcast 0.1 1.56 0.19 2.82 0.22 3.31 0.22 3.31

Scalable 1.17 16.30 1.09 15.33 1.25 16.70 1.25 16.70