56
J. Eur. Opt. Society-Rapid Publ. 21, 36( 2025)
Table
1. Hasler colorfulness values of the images shown in Figure 3 and used for the subjective quality evaluation.
Images |
Image colorfulness range |
( u) |
Between 0 and 0.05 |
( c)-( o)-( s) |
Between 0.05 and 0.10 |
( d)-( g)-( i-j)-( q)-( t) |
Between 0.10 and 0.15 |
( a-b)-( l)-( r) |
Between 0.15 and 0.20 |
( m)-( p) |
Between 0.20 and 0.25 |
( f)-( v) |
Between 0.25 and 0.30 |
( e)-( w) |
Between 0.30 and 0.35 |
( h) |
Between 0.35 and 0.40 |
( n)-( x) |
Between 0.40 and 0.45 |
enforced, participants were instructed to perform the experiment in a well-lit room without direct glare on the screens and to avoid extreme ambient lighting. All observers used their computer rather than their mobile phone, whose resolution is assumed to be at least 1920 1080. To mitigate viewing variability, relative comparisons( rather than absolute color judgments) were used throughout the test design. The observers were asked to rank the images from 1 to 6, with 1 representing the closest match to the original image and 6 the least faithful reproduction according to them. The observers utilized their own display devices and were instructed to ensure that the night filter was disabled and that the screen exhibited sufficient luminosity. As the images were ranked in relative order and not absolute order there was no need to use color calibrated displays.
3 Results
A total of 34 observers participated in this experiment, from nine different countries, primarily in Europe. The distribution of answers is illustrated in the histograms presented in Figure 4. This figure displays 24 histograms, each corresponding to 1 of 24 images presented in Figure 3. Each histogram depicts, along the y axis, the number of individuals who ranked the image simulated with each gamut( corresponding to different colors) from“ rank 1” to“ rank 6”. The x-axis represents the rank assigned to each simulated image. The pink bar is for the portrait gamut, cyan for the printer one, purple for the bad one, orange for“ 1 _ BR”, green for“ 6 _ BR” and“ 8 _ BR” for the red one. The height of each color segment within a bar reflects the number of observers who assigned that particular gamut to the given rank, making it possible to visualize ranking tendencies across images.
As evidenced by the histograms, the question posed was effectively understood by the observers. The intentionally bad gamut was consistently ranked last by a large majority of observers, indicating that there were no inversions in ranking. In all observations, no responses were identified as anomalous, indicating that filtering was unnecessary. This was confirmed by the Z-score method [ 37 ], which filters all values exceeding two standard deviations from the
Figure 4. Distribution of preference rankings for each image(( a)–( x)) from Figure 3 based on the overall ranking. The x-axis represents the assigned rank, and the y-axis indicates the number of observers who assigned each rank.
mean, and did not change the results. Furthermore, as illustrated in Figure 4, the experimental gamuts appear to perform better than the intentionally bad gamut, and worse than the printer and portrait gamuts for almost all observers.
With respect to the ranking from the observers( Fig. 4), the portrait gamut is ranked first for the portrait images( Figs. 3a – 3d, 3i – 3l), in accordance with expectation, as well as for images( 3g, 3o – 3u), exhibiting a Hasler colorfulness value of less than 0.25 as shown in Table 1. In the case of images exhibiting a Hasler colorfulness exceeding 0.25, the printer gamut is preferred by observers, despite its lower contrast, as evidenced by the responses to images 3e and 3h. The results are less conclusive for the three experimental gamuts( orange, green and red bars) that the observers ranked between ranks 3 and 5, with a ranking rarely occurring in the same order.