60
J. Eur. Opt. Society-Rapid Publ. 21, 36( 2025)
Figure 5. Ranking resulting from the subjective IQA evaluation( Pref) vs predicted ranking based on equations( 3) and( 4)( Pred). The colors correspond to the different images.
Figure
7. Actual and predicted ranks for each image.
Figure
8. Confusion matrix between actual and predicted ranks for all images.
Figure 6. Simulated images with the six gamuts and comparison of predicted and subjective rankings.
artificial good and bad gamuts from the experimental, but there is still confusion to rank the experimental gamuts reproduction.
To address the observed ambiguity in subjective rankings across experimental gamuts mainly, a plot of the actual and predicted ranks is shown in Figure 7. In this figure, vertical error bars represent a ± 0.7 rank tolerance which was the average standard deviation among the observers’ ranking for the experimental gamuts. This threshold accounts for minor perceptual differences that may not be reliably distinguished by observers. Points whose error bars intersect the diagonal( y = x) are marked in green, indicating that the predicted ranking falls within an acceptable range of the true subjective rank. This visual representation highlights that 138 out of 144( and 69 out of 72 for experimental gamuts) misalignments are minor and localized.
To further assess the consistency between the predicted and actual subjective rankings, a confusion matrix( Fig. 8) was introduced using the ordinal ranks of each gamut per image. This representation makes it easier to visualize correct matches and mismatches between predicted and observer-derived rankings. For 15 out of the 24 tested images, all six gamuts were ranked in the same order as those provided by human observers. Among the remaining cases, the mismatches were limited to close gamuts, suggesting localized ambiguity rather than a broader failure in prediction. Notably, these discrepancies primarily