A visual neuron doesn't report brightness — it reports contrast. Its centre says 'light here', its surround says 'but not there', and the difference sharpens every edge into a bright and dark band that isn't really in the world. Drag the edge, slide the surround, flip the polarity, and watch the Mach bands appear.
Each neuron's receptive field is a Difference-of-Gaussians: a narrow excitatory centre minus a broad inhibitory surround (drag the surround strength). The response at every point is the luminance convolved with that kernel — computed live across the row. Slide an edge (dark→light) through it and the response overshoots just inside the bright side and undershoots just inside the dark side: Mach bands, the illusory bright/dark stripes at edges. Flip to OFF-centre and the whole response inverts. Real lateral-inhibition / centre-surround vision.
A one-dimensional row of identical Difference-of-Gaussian cells stands in for a real retina/LGN (which is 2-D, has many cell sizes, adapts, and adds noise and nonlinearity). The mechanism shown — centre-surround convolution producing edge enhancement and Mach bands — is the exact, real model; the specific kernel widths are illustrative.