Image Processing Reference
In-Depth Information
Table 5.2 BD-PSNR and BD-BR test results for Laura
3D Holo Scalable
(PSNR [dB] and BR [%])
Basic
Rendering,
patch size 10
Weighted
Blend, patch
size 10
Basic
Rendering,
patch size 20
Weighted
Blend, patch
size 20
PSNR BR
PSNR BR
PSNR BR
PSNR BR
HEVC Simulcast
2.26
30.05
2.24
29.87
2.52
32.58
2.23
29.73
3DHolo Simulcast
0.51
0.26
4.05
0.06
0.03
0.01
0.23
0.00
Scalable
2.03
27.76
2.12
28.61
1.87
25.97
2.16
29.07
a
b
Weighted Blend, with patch size of 4
Basic Rendering, with patch size of 4
44
42
40
38
36
34
44
42
40
38
36
34
HEVC Simulcast
HEVC Simulcast
Scalable
Scalable
3DHolo Simulcast
3DHolo Simulcast
32
32
3DHolo Scalable
3DHolo Scalable
30
30
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
[bpp]
[bpp]
c
d
Basic Rendering, with patch size of 10
Weighted Blend, with patch size of 10
44
42
40
38
36
34
44
42
40
38
36
34
HEVC Simulcast
HEVC Simulcast
Scalable
Scalable
3DHolo Simulcast
3DHolo Simulcast
32
32
3DHolo Scalable
3DHolo Scalable
30
30
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
[bpp]
[bpp]
Fig. 5.16 RD performance of Scalable Codec for Plane and Toy
compared to the independent compression of the holoscopic content in relation to
the content in the 2D and multiview layers (as discussed in Sect.
5.4
). Thus, only the
PSNR and bitrate values of the
Second Enhancement Layer
are considered in the
RD charts.
Based on these results, it can be seen that the presented display scalable coding
scheme,
3DHolo Scalable
, always outperforms the simulcast solution based on the
HEVC (
HEVC Simulcast
). Furthermore, as can be seen in Fig.
5.16
, for
Plane and
Toy
test image, it is possible to take advantage of performance benefits of both
prediction methods (self-similarity and inter-layer prediction) to improve the over-
all performance, since the
3DHolo Scalable
scheme always outperforms the sce-
nario where each prediction method is used alone (i.e.,
3DHolo Simulcast
and
Scalable
).