Quality Assessment - 3DTV: Processing and Transmission of 3D Video Signals - page 182

Table 6.5 Performance of overall depth assessment models for
3D video sequences

Objective quality model Depth quality

CC RMSE SSE

Average PSNR of the Rendered 0.7788 0.07375 0.0579

Left and Right views
Average SSIM of the Rendered 0.8065 0.06745 0.05478

Left and Right views
Average VQM of the Rendered

0.7753 0.07397 0.0603

Left and Right views
Proposed Depth Quality Model

0.8716 0.03253 0.03791

Since a VQM score of 0 implies the best quality and 1 implies the worst
with respect to the original video, D M in Equation (6.9) is defined as 1- VQM .
Subjective ratings for depth quality from test 3 are used to assess the
performance of the proposed model. Average PSNR, SSIM and VQM qual-
ity ratings of rendered left and right video are used as measures of depth
quality for performance comparison of the proposed model. The relationship
between the MOS for perceived depth quality and the assessment models are
approximated by a symmetrical logistic function (6.3). The performance com-
parison metrics (CC, SSE, RMSE) for each prediction model, approximated
using the symmetrical logistic function, are evaluated for all test sequences
and results are presented in Table 6.5. α and β in Equation (6.9) are varied

from 0 to 5 in steps of 0.5 and 1 revealed the best correlation

with the subjective ratings. However, the optimum values of

α = 1.5, β =

will
vary for different types of displays. Results imply that-the proposed model
has better values for the performance comparison metrics, with regards to
subjective ratings in predicting depth quality of colour-plus-depth-based 3D
video. Therefore, the proposed model, based on visually important features
to the brain (both monocular and binocular), can be used to predict depth
quality of 3D video.

α and β

6.4.3 Compound3DVideoQualityModel

The next task is the design of a compound 3D video quality model by combin-
ing the image and depth quality models proposed earlier. In general, when
different user (observer) groups and terminal characteristics are considered,
the relative importance of image and depth quality of 3D video depends
on content characteristics, context characteristics (display characteristics and
lighting conditions) and user preference. However, in this study, effects
of different context characteristics and preferences of the 3D observer are
not considered. Thus, the compound 3D quality measure is modelled with