A unifying principle of chromatic coding across biological and artificial systems

Abstract

Color is a defining feature of human vision, yet its integration with spatial structure across stages of the visual system is still not fully understood. Classical accounts assumed that color provides little spatial information, being represented coarsely and separately from luminance. Here, we show that the spatial selectivity of color is not fixed, but dynamically transforms with temporal frequency. In human observers, steady-state visual evoked potentials reveal a clear shift from low-pass tuning at higher temporal frequencies to band-pass tuning at lower frequencies. Local field potentials recorded from macaque V1 exhibit the same transition, and color-deficient observers show a selective loss of the low-pass component, pointing to distinct underlying mechanisms. Analyses of deep neural networks trained for object recognition reveal an analogous transformation, demonstrating that this principle also emerges in artificial vision systems. Together, these findings establish that the spatial tuning of color evolves systematically with temporal scale, providing a unifying principle of chromatic coding across biological and artificial systems.