Jin Hirano¹, Masataka Sawayama^2,3, Kiyofumi Miyoshi¹, Shin'ya Nishida¹; ¹Kyoto University, ²Hokkaido University, ³Prometech CG Research, Japan

Large individual differences in color perception, as observed with #TheDress and #TheShoe, can be attributed to the ambiguity in specifying the surface–illuminant relationship arising from the limited chromatic information in images. One factor that may increase the ambiguity in these images, and thereby increase individual differences, is that they have approximately one-dimensional chromaticity distributions (elongated along the daylight locus or along two complementary colors). According to this idea, making the chromaticity distribution two-dimensional will reduce inter-observer variability in perceived colors. To test this, we psychophysically examined how the dimensionality of the chromaticity distribution affects inter-observer variations in categorical color judgments. We created computer-rendered cube stimuli presented against a black background. Each face of the cube contained a 6×6 grid of tiled surface colors. We manipulated the dimensionality of the stimulus chromaticity distributions by changing the number of surface colors used across the grid tiles. Stimuli with one and two colors formed one-dimensional distributions, while those with four colors and multiple colors exhibited two-dimensional distributions. To systematically vary color appearance, we shifted the chromaticity distribution along a color axis in the CIE L*u*v* space. The experimental procedure followed Hirano et al. (VSS2025). For each participant, we estimated the color switching point at which each participant’s perception switched between alternative interpretations (e.g., “white and gold” to “blue and black”). Inter-observer variations in switching points were substantially larger for the two-, four-, and multi-color conditions than for the one-color condition. Contrary to the expectation, the two-, four-, and multi-color conditions yielded comparable levels of variability. The color distribution along the daylight-locus axis tended to show slightly greater variability than that along the other axes. This result suggests that the one-dimensional chromaticity distribution is not a determining factor in explaining inter-observer variability in color perception.

Acknowledgements: This work was supported by JST SPRING (JPMJSP2110) to JH and JSPS KAKENHI425 (JP20H05957, JP24H00721) to SN.