Environmental color statistics shape the anisotropic geometry of human color discrimination
Hedjar, L.; Morimoto, T.; Akbarinia, A.; Bartsch, M. V.; Strumpf, H.; Hopf, J.-M.; Gegenfurtner, K. R.
Show abstract
Hue is a powerful cue for object discrimination, but human color discrimination is anisotropic: hue thresholds are lower than chroma thresholds for orangish colors, but nearly equal for purplish colors. Here we show that this asymmetry reflects environmental color statistics rather than being a fixed consequence of cone-opponent architecture. Across fifteen image and reflectance databases, orangish hues accounted for 62.5% of chromatic samples, compared with 5.3% for purplish hues. We replicated the hue-chroma asymmetry psychophysically in 44 participants. Magnetoencephalography revealed a corresponding neural asymmetry, with superior decoding of orangish hue differences emerging around 250 ms after stimulus onset. Deep neural networks trained on naturalistic image datasets reproduced the human-like asymmetry without explicit color supervision. Critically, training on hue-inverted images reversed the asymmetry, producing greater hue sensitivity for purplish than for orangish colors. These results suggest that the geometry of color discrimination is shaped by ecological chromatic structure.
Matching journals
The top 4 journals account for 50% of the predicted probability mass.