Visual System Lecture: Retina to V1 – Key Concepts and Mechanisms

Rods are absent from the fovea, while cones reach their highest density there and gradually thin toward the periphery. Cones dominate vision under bright (photopic) conditions; as illumination drops, rods—more sensitive to light—take over, shifting perception to a rod‑dominated mode.
Ganglion cells form the retina’s output layer, sending axons through the optic nerve. Two major types, midget and parasol cells, differ in receptive‑field size and functional role. Receptive fields expand with eccentricity, so spatial resolution falls off sharply outside the fovea.

Computational Models of Vision

The fovea represents an optimized solution for high‑resolution sampling when the optic nerve provides a limited number of fibers. Machine‑learning simulations of visual‑search tasks develop fovea‑like receptor lattices when constrained to translation, confirming the efficiency of this arrangement.
Convolution offers a convenient way to model the response of a population of neurons that share a receptive‑field shape but occupy different retinal locations. In these models, each neuron behaves as a linear filter whose output passes through a non‑linearity, reproducing many observed visual‑system behaviors.

Visual Illusions and Mechanisms

Center‑surround receptive fields compute the difference between illumination in the central region and the surrounding annulus. On‑center cells increase firing for luminance increments in the center, while off‑center cells respond to decrements.
Mach bands, grid, and intersection illusions often arise from these center‑surround filter responses. The explanations assume the brain does not “know” the source of the filter output, so the visual system can be tricked by unnatural, high‑contrast stimuli. As one lecturer put it, “Essentially, these explanations implicitly assume suboptimal decoding of the responses, which is kind of weird.”

Central Visual Pathways

Visual information from the left visual hemifield projects to the right lateral geniculate nucleus (LGN) and vice versa. Within the LGN, six layers segregate inputs by eye and by cell type: the four outer layers are parvocellular, handling color and fine form, while the two deeper layers are magnocellular, specialized for motion and flicker.
Lesion studies using ibotenic acid in macaque monkeys demonstrate that destroying parvocellular layers impairs color and pattern discrimination, whereas magnocellular lesions disrupt motion perception.

Primary Visual Cortex (V1)

Cortical magnification allocates a disproportionately large cortical area to the fovea, compressing peripheral representations. This non‑uniform mapping preserves high‑resolution sampling where it matters most.
Neurons in V1 exhibit orientation selectivity, responding preferentially to lines of a particular angle. Hubel and Wiesel showed that this selectivity likely emerges from specific wiring of LGN inputs with center‑surround receptive fields. Orientation columns group neurons with similar preferences, forming smooth maps across the cortex.
Population coding resolves ambiguities that arise when a single neuron’s response could reflect multiple stimulus attributes (e.g., contrast versus orientation). By examining the peak response across a neuronal ensemble, the visual system disambiguates the stimulus.
Adaptation further refines perception: prolonged exposure to a specific orientation reduces the responsiveness of corresponding neurons, producing the tilt aftereffect, where a subsequently viewed vertical line appears tilted in the opposite direction. As noted in the lecture, “Adaptation does happen. And what we will talk about when we resume next time is how we can explain what just happened to you in terms of adaptation and population codes.”