Image Processing Reference
In-Depth Information
the HVS will not be able to fuse the affected views, and this causes binocular
rivalry. This has detrimental effects on the final QoE perceived by the end user.
Recent studies on 3D video transmission [
4
] have found that binocular rivalry is
influencing the overall perception and this effect prevails over the effect of binoc-
ular suppression. To avoid the detrimental effect of binocular rivalry, the transmis-
sion system could be designed appropriately taking this issue into account. For
instance, the transmission system parameters can be updated “on the fly” to obtain
3D views with minimum distortions, according to the feedback on the measure of
3D video quality at the receiver-side. In case of low quality due to different errors in
the two views, if the received quality of one of the views is significantly low, the
transmission system could be informed to allocate more resources to the worse view
or to increase the error protection level for that 3D video channel to mitigate the
quality loss in subsequent frames. This increases the opportunity to fuse the 3D
video content more effectively and improve the final QoE of users. The measured
image quality at the receiver-side can be used as feedback information to update
system parameters “on the fly” in a “QoE-aware” system design approach as
discussed in [
2
]. However, measuring 3D video quality is a challenge mainly due
to the complex nature of 3D video quality [
7
] and the necessity of sending a copy of
the original 3D image sequence for measuring the quality with FR methods.
Immersive video quality evaluation is a hot topic among researchers and devel-
opers at present, due to its complex nature and to the unavailability of an accurate
objective quality metric for 3D video. 3D perception can be associated with several
perceptual attributes such as “overall image quality,” “depth perception,” “natural-
ness,” “presence,” “comfort,” etc. A detailed analysis is necessary to study how
these 3D percepts influence the overall perceived 3D image quality in general. For
instance, the study presented in [
8
] concludes that excessive disparities between left
and right view can cause eye strain and therefore degrade the perceived image
quality. Mostly, appreciation-oriented psychophysical experiments are conducted
to measure and quantify 3D perceptual attributes. A few standards also define
subjective quality evaluation procedures for both 2D and 3D video (e.g., [
9
,
10
]).
However, these procedures are not competent enough to measure 3D QoE param-
eters and show several limitations; for instance these are not able to measure the
combined effect of different perceptual attributes. Subjective quality evaluation
under different system parameter changes have been reported in a number of studies
[
8
,
11
,
12
]. However, these studies are limited to certain types of image artifacts
(e.g., compression artifacts) and have limited usage in practical applications. On the
other hand, subjective quality evaluation requires time, effort, controlled test
environments, money, human observers, etc. and cannot be deployed in a live
environment where quality is measured “on the fly.”
Objective quality evaluation methods for 3D video are also emerging to provide
accurate results in comparison to the quality ratings achieved with subjective tests
[
13
,
14
]. However, the performance of these metrics is most of the time an approx-
imation to that of subjective quality assessments. Our recent studies have also found
out that there is a high correlation between subjective ratings and individual objective
quality ratings of 3D video components (e.g., average PSNR and SSIM of left and
