Quality of Experience for High Definition Presentations – Case: Digital Cinema - High-Quality Visual Experience - page 42

blocks, while the remaining blocks with lower contrast levels were excluded from
the quality computation.

Table 3 gives the evaluation results of the original metrics and the above two
methods. According to the evaluation results, the proposed approach can evidently
improve the performance of the original metrics in image quality assessment of
digital cinema, especially for SSIM. The first method can improve the perform-
ance of these metrics, especially for SSIM and PHVS, which indicates subjects
formulate their judgment on the image quality in a digital setup based on a all
blocks, rather than the global image. Further, the second method has better per-
formance when compared to the first method. It indicates that subjects pay more
attention to those image regions with higher contrast levels when assessing the
image quality. This observation will be useful to develop a quality metric for digi-
tal cinema applications.

Table 3 Evaluation results of different methods on image quality assessment in digital
cinema setup

Original metrics First method Second method

Criteria

PSNR SSIM PHVS PSNR SSIM PHVS PSNR SSIM PHVS

RMSE 1.00 1.13 0.85 1.00 1.00 0.77 0.73 0.53 0.56

Pearson 0.914 0.888 0.938 0.914 0.913 0.949 0.954 0.976 0.974

Spearman0.913 0.875 0.941 0.913 0.904 0.952 0.956 0.974 0.976

Outlier
ratio

0.063 0.063 0.021 0.042 0.063 0.021 0.021 0.021 0

Table 4 Evaluation results of Minkowski summation for spatial pooling

Criteria PSNR SSIM PHVS
RMSE 0.71 0.54 0.57
Pearson 0.958 0.976 0.973
Spearman 0.961 0.974 0.976
Outlier ratio 0.021 0.021 0.021

In the above experiments, we used the direct mean of quality values over the
blocks. In addition, another pooling scheme, Minkowski summation, is widely used
in some quality metrics to pool the quality values over different spatial regions,
such as the perceptual distortion metric (PDM) proposed by Winkler et al. [32].
Therefore, we also tested the Minkowski summation with different exponents on
those blocks with higher contrast levels. The experimental results demonstrated that
the best performance was achieved when the exponent in the Minkowski summa-
tion was set to 2, and the results indicated that the Minkowski summation can also
be used in pooling the quality values over different blocks in the proposed
approach, as shown in Table 4.