J. Eur. Opt. Society-Rapid Publ. 21, 36( 2025) 61
Table 6. Predicted and subjective ranking for each gamut averaged over the 24 images. Gamut
Average predicted ranking
Standard deviation predicted ranking
Average subjective ranking
Standard deviation subjective ranking
Portrait |
1.42 |
0.22 |
1.33 |
0.28 |
Printer |
1.98 |
0.38 |
1.94 |
0.42 |
8 _ BR |
3.78 |
0.64 |
3.83 |
0.91 |
6 _ BR |
3.83 |
0.44 |
3.98 |
0.53 |
1 _ BR |
4.19 |
0.52 |
4.20 |
0.81 |
Bad |
5.60 |
0.13 |
5.72 |
0.03 |
Table 7. Hasler colorfulness values of images shown in Figure S1.
Image |
Hasler colorfulness metric |
( a) |
0.16 |
( b) |
0.15 |
( c) |
0.09 |
( d) |
0.15 |
( e) |
0.31 |
( f) |
0.25 |
( g) |
0.12 |
( h) |
0.35 |
stemmed from confusion between two gamuts that often received similar rankings. Appendix D shows the observer ranking variability in the case of a difficult to rank image.
As it looks in Table 6, while the metrics do not always perfectly distinguish the best gamut among the actual gamuts, they still reveal meaningful patterns. This aligns with the observed variability in subjective rankings, where the standard deviation was notably high for the experimental gamuts.
The proposed linear models were initially calibrated for reflective structural gamuts; the generalization potential of these models to other imaging systems requires further consideration. While the current models provide an effective framework for reflective displays, their applicability to emissive or transmissive displays would need to be reevaluated, especially in cases where the perceptual weighting of metrics may differ between systems, despite the models being robust to low variability on observers ranking( Appendix E).
The key metric most likely to be affected by a change in illuminant is the“ Distance from the selected white point to the image gamut”( M16). This metric quantifies how a shift in the illuminant alters the perceived white point in the color palette, which is essential for evaluating color reproduction under varying lighting conditions and across different display technologies. Therefore, the applicability of the linear models should be reassessed when considering such changes in illuminant, ensuring their robustness in diverse imaging contexts.
The metrics seems to be usable for transmission mode colors for the samples to some degree as shown in Appendix B.
This metric has the interest of being linked not only to the gamut used for printer reproduction but also to the original image considered.
5 Conclusion
In this study, we established a relationship between several objective metrics from the literature and a subjective IQA of images simulated with different color gamuts using an observer-ranked method.
The relative consistency of the study results suggests that the observers had a clear understanding of the ranking protocol, but some difficulties to rank experimental gamuts that have different kinds of limitations. We identified a combination of metrics that efficiently differentiate printer reproduction quality when the variations are sufficiently pronounced. However, the model’ s performance proved less reliable for experimental gamuts, as participants encountered difficulties ranking these reproductions despite having access to the original image, highlighting the inherent subjectivity of the task.
Moving forward, this approach could be extended to other observation modes, as the colors studied here correspond to the ones printed by a laser-induced printer on transparent plasmonic films, observed on backside reflection mode, which are significantly different than those on transmission and frontside reflection modes. Adding observers’ data from images simulated on those modes could add some robustness to the model. The results of this study, along with the extracted metrics, provide valuable insight into selecting the most suitable gamut for printing a given image. This, in turn, contributes to a better understanding of gamut quality and its impact on the final printer reproduction.
Acknowledgments
The authors thank the observers for giving their time for this research by answering the surveys.
Funding
This work is funded by the ANR project SLICID( ANR-23-CE39- 0006).