V1 Cortical Maps

orientation preference, ocular dominance, and why they are perpendicular

The primary visual cortex (V1) is organized into two overlapping maps. The orientation map records each neuron's preferred stimulus angle — it forms smooth pinwheel patterns where all orientations radiate from a central singularity. The ocular dominance map records which eye drives a neuron more strongly — it forms alternating stripes, one per eye.

Both maps can be modelled as Gaussian random fields whose Fourier spectra sit on a narrow ring of spatial frequencies. The remarkable structural relationship: if the OD frequency ring is the orientation ring rotated by 90°, the two maps' iso-contour lines cross at right angles everywhere. This is the heart of Hubel & Wiesel's ice cube model.

Below, the left panel shows the Fourier domain. The yellow dots are the sampled orientation frequencies; the cyan dots are the same frequencies rotated 90°, used for the OD map. Drag the polygon vertices to reshape the frequency support and watch both maps update in real time. Toggle gradient ticks to see the local gradient directions — they are perpendicular between the two maps.

interactive — drag polygon vertices in the Fourier domain to reshape both maps

Fourier domain · drag white vertices

orientation preference map

−180°0°+180°

ocular dominance map

left eyebalancedright eye

number of waves 32

frequency scale 10.0

low → coarse columns · high → fine columns

The yellow ticks on the orientation map show the orientation gradient direction. The cyan ticks on the OD map show the orientation gradient direction too — notice they always cross the OD stripe boundaries at right angles.

Three maps at 120°

orientation, ocular dominance, and spatial frequency — and why they cannot all be perpendicular

Two maps can sit at 90° to each other. But V1 carries a third periodic feature: spatial frequency. Three independent maps cannot all be mutually perpendicular — a 2D cortical sheet only has two dimensions to spread across, so three orthogonal axes will not fit. The maximally-even compromise places the three frequency-support polygons 120° apart.

The Fourier panel below shows one polygon (orientation) plus two copies of it rotated by 120° (ocular dominance) and 240° (spatial frequency). Drag any orientation vertex — the other two polygons rotate rigidly with it, so the 120° relationship is preserved while you reshape the support. Each map is a Gaussian random field built from its own rotated copy of the same weights, so the three maps are siblings differing only by rotation.

interactive — the three frequency polygons are locked 120° apart; drag a yellow vertex to reshape all three

Fourier domain · three polygons at 120°

orientation preference

−180°0°+180°

spatial frequency preference

low SF high SF

number of waves 36

frequency scale 10.0

ocular dominance

leftbalancedright

Each map carries gradient ticks in its own colour. Compare any two maps: their dominant gradient directions differ by 120°, not 90°. With three features competing for the same sheet, perfect orthogonality is geometrically impossible — 120° is the fair split.

This is an idealization. Real cortical map spectra are blurry annuli with substantial angular spread, broken by fractures and pinwheels, and the measured relationship between ocular dominance and spatial frequency is contested — orthogonal in some studies, parallel-with-distinct-periods in others. The 120° arrangement is the clean model of the coverage constraint, not a measurement. Retinotopy, the fourth map, does not appear on these rings at all: it is a single half-cycle across the whole sheet, so its power collapses to the origin.