Page 146 - Proceedings of the 2017 ITU Kaleidoscope
P. 146
2017 ITU Kaleidoscope Academic Conference
6. RESULTS AND DISCUSSION
The proposed approach SBQA-SP and its variant SBQA-
GP along with existing QBQA algorithms were
implemented and tested in typical internet environment.
The system performance was evaluated using full reference
video quality metrics: PSNR, SSIM, MS-SSIM and VQM.
The numerical data representing different quality index
arising out of live streaming of video were analyzed offline.
6.1. Peak Signal to Noise Ratio (PSNR)
The commonly usedvideo / image quality metric PSNR is
used here because it is fast and simple to implement.Figure Fig. 4. The SSIM index
3depicts the PSNR values corresponding to the SBQA-SP,
SBQA-GP and QBQA algorithms. The SBQA-SP 6.3. Multi Scale Structural Similarity (MS-SSIM)
algorithm exhibits a higher average PSNR, which is 8% and Measurement
5% higher than the SBQA-GP and QBQA algorithm
respectively. The SBQA-SP performs better compared to A multi scale SSIM being more flexible than single scale
other two approaches as it learns the environment metric provides better result with respect to correlations to
conditions without any pre learning phase and adapts the human perceptions. On an average SBQA-SP algorithm
video quality as data arrives. Also, the exploration policy produces 1.2% higher quality on MS-SSIM scale than
used helps to convergence at a faster rate.
QBQA algorithm and 2.2% SBQA-GP algorithm. Figure 5
shows the variation of MS-SSIM values on different
consecutive frames. Although the improvement in quality
Fig. 3. The PSNR observation
6.2. Structural Similarity Measurement (SSIM) Fig. 5. The Multi Scale SSIM index
The SSIM index was computed for the three algorithms bySBQA-SP algorithm over other two is marginal, it still
(SBQA-SP, SBQA-GP and QBQA), and the proposed leads the pack.
SBQA-SP algorithm exhibited a higher value with an
average index of 0.9940 which is 2% higher than the 6.4. Video Quality Metric (VQM)
QBQA algorithm 3% better than SBQA-GP. The SSIM
index on few decoded consecutive frames at receiver The VQM is an important metric for the evaluation of video
corresponding to three algorithms is plotted in Figure4. The quality because it considers the spatial-temporal aspects of
higher SSIM index for the proposed algorithm is a reward visual perception. Since the VQM score is the sum of many
for the perceived video quality as there is need for the weighted parameters and its higher value represents the
perseverance of luminance and contrast factors that are maximum loss of quality in the video, the lower values
influenced by the distortions. observed by the SBQA-SP, SBQA-GP and QBQA
algorithms are desirable. The SBQA-SP based method
shows 15% lower than the SBVQA-GP and 23% lesser than
the QBQA algorithm. The VQM values for 20 Frames are
depicted in Figure 6.
– 130 –