Analysis of Variance of Opinion Scores for MPEG-4 Scalable and Advanced Video Coding

Abstract

In this paper, we conduct an analysis of variance (ANOVA) on opinion scores for MPEG-4 scalable video coding (SVC) and advanced video coding (AVC) standards. This work resorts on a publicly available database providing opinion scores from subjective experiments for several scenarios such as different bit rates and resolutions. In particular, ANOVA is used for statistical hypothesis testing to compare two or more sets of opinion scores instead of being constrained to pairs of sets of opinion scores as would be the case for t-tests. As the ANOVA tests of the different scenarios are performed for mean opinion scores (MOS), box plots are also provided in order to assess the distribution of the opinion scores around the median. It is shown that the opinion scores given to the reference videos in SVC and AV C for different resolutions are statistically significantly similar regardless of the content. Further, for the opinion scores of the considered database, the ANOVA tests support the hypothesis that AVC generally outperforms SVC although the performance difference may be less pronounced for higher bit rates. This work also shows that additional insights on the results of subjective experiments can be obtained by extending the analysis of opinion scores beyond MOS to ANOVA tests and box plots.