Difference-of-Gaussians (DoG) model
RF(r) = Kc·exp(−r²/2σc²) − Ks·exp(−r²/2σs²)
where σc = center radius, σs = surround radius.
On-center cells: fire strongly to small bright spots; surround inhibition cancels large uniform fields (explains "Mach bands" and Hermann grid).
Spatial frequency tuning: the DoG acts as a bandpass filter — maximum response at f = 1/(2σc). Move the cursor on the RF diagram to probe response.
Lateral inhibition creates edge enhancement.
The retinal output (convolution of input with DoG kernel) sharpens luminance edges — the physical basis of Mach bands. Each row shows (top→bottom):
1. Input luminance profile
2. DoG kernel
3. Retinal output (filtered)
The overshoot near the edge is neurally real — perceived brightness tracks the filtered signal, not raw luminance.