Real-Time 3D QoE Evaluation of Novel 3D Media - Novel 3D Media Technologies - page 173

In the second case, transmission over a wireless channel is simulated considering
an OFDMA based transmission system (such as the recent WiMAX and 3GPP LTE
systems). All the layers of the OSI protocol stack are realistically simulated, from
packetization to transmission over the network through RTP/UDP/IP and over the
MAC/Physical layer. The wireless channel is simulated through a Rayleigh block
fading channel in time and frequency, with block duration of 0.1 s and channel
coherence bandwidth of 1.25 MHz. In addition to fast fading, log-normal block
fading is taken into account, in order to model slowly varying shadowing effects
( ˃ [ dB ] ¼8 dB and block duration of 8 s). The 3D video clients are characterized by
different median signal-to-noise ratios ( E S / N 0 ). LDPC codes are adopted, together
with Adaptive Modulation and Coding, with a target frame error rate of 10 2 .
In all cases, “ Slice copy ” is enabled during decoding to conceal the missing
slices of the corrupted bit-stream. At each PLR /( E S / N 0 ) and QP value, the quality is
measured using the FR SSIM and the proposed methods for depth map and
corresponding color images.
In order to show the degree of correlation between the proposed quality ratings
and the true 3D video perception, subjective tests are performed using the double
stimulus impairment scale (DSIS) method with the participation of 16 observers.
47” LG display with polarized glasses is utilized for these tests. Processed color and
depth map sequences are rendered into left and right views using the DIBR method
according to MPEG informative recommendations [ 36 ]. The dis-occluded regions
(visual holes) are filled by background pixel extrapolation method, described in
[ 37 ]. Standard test room under normal room lighting condition is selected for the
tests. Subjects are screened for visual acuity (using the Snellen chart), color
blindness (using the Ishihara test), and stereo vision (using the Butterfly Stereo
test) prior to the tests. Two training sessions are conducted to familiarize with the
3D display and the test sequences used in the subjective tests. Subjective quality
(i.e., overall 3D video quality) is measured for all the sequences under different
PLR s. The sequences generated with QP ¼

10 and QP

¼

30 are selected for these
tests. The opinion scores ( OS ) of individual subjects are averaged to obtain mean
opinion score ( MOS ) for all the test cases. In addition to subjective quality ratings,
the relationship between the proposed method and average PSNR quality of ren-
dered left and right views is also demonstrated. This will also show the degree of
correlation between the proposed quality ratings and the quality rating for rendered
views using DIBR.
Figure 9.5 shows the quality measured with the proposed and SSIM quality
metrics for color and depth map images of the Ballet sequence at QP 30 (Fig-

ure 9.5a color image quality @ all PLRs and Fig. 9.5b depth map quality @ all
PLRs). It can be clearly seen that both measures are closely matched for both color
and depth map quality measurements. In order to minimize the offset between the
proposed and the FR quality ratings, a relationship can be derived between the FR
(SSIM D and SSIM C ) and the proposed methods (i.e., E-SSIM C and E-SSIM D ) based
on experimental data using ( 9.12 ) and ( 9.13 ).

¼