Image Processing Reference
In-Depth Information
FR quality ratings of the individual color (SSIM C ) and depth map (SSIM D ) images.
Equations ( 9.12 ) and ( 9.13 ) define the relationships between the proposed and FR
methods.
SSIM C
f Colour E-SSIM C
ð
Þ
ð
9
:
12
Þ
SSIM D
f Depth E-SSIM D
ð
Þ
ð
9
:
13
Þ
where SSIM C and SSIM D are FR SSIM maps of color and depth map images
respectively.
Even though the proposed quality estimation obtained with Eqs. ( 9.12 ) and
( 9.13 ) provides SSIM ratings for color and depth map sequences separately, it
may not provide direct relationship between these measures and subjective 3D
video quality results. This is mainly due to the nature and usage of color and depth
map sequences in 3D view rendering (i.e., Depth map is not directly viewed by the
users and only used for 3D video rendering.) However, the existing relationship
between these objective measures and subjective quality ratings (e.g., overall 3D
image quality, depth perception) are studied in our previous research [ 15 - 17 ]. As
elaborated in Section 9.1 , these studies show a strong correlation between subjec-
tive and objective quality ratings (including the SSIM metric) for a range of
compression rates and PLR s. In this chapter also we provide experimental results
to show the degree of correlation between individual objective quality measures
and true 3D subjective quality ratings obtained after series of subjective quality
evaluation tests.
9.3 Experimental Setup, Results, and Discussion
In order to evaluate the performance of the proposed NR quality metric for color
plus depth 3D video, experiments are performed for different PLR s (i.e., 0, 3, 5, 10,
and 20 %) and compression levels (i.e., with different QP values). The Orbi ,
Interview , Ballet , and Breakdance 3D video test sequences are encoded using the
H.264/AVC video coding standard ( JM reference software Version 16.0).
Ten-second long sequences (i.e., 250 frames from Orbi and Interview sequences-
25 fps and 150 frames from Ballet and Breakdance sequences-15 fps) are encoded
with IPPPIPPP
format, using QP values 1, 5, 10, 15, 20, 25, 30, 35, 40, 45, and
50. An I frame is encoded by every one second. Slices (one row of MB s
...
one slice)
are also introduced in order to make the decoding process more robust to errors.
The transmission of the encoded bit-stream is simulated over an IP core network
and over an LTE-like wireless system. In the first case, transmission is simulated by
using IP error patterns generated for Internet experiments [ 30 ]. In order to obtain
average results, random starting positions are used for the error pattern files. The
corrupted bit-streams with different PLR s are later decoded using JM reference
software decoder.
¼
Search WWH ::




Custom Search