Concept

Vision Processing

Intro

What lands on your retina is a small, upside-down, two-dimensional, jittering blur, split across two eyes with a blind spot in each. What you experience is a stable, upright, seamless, three-dimensional world you can move through and act on. Between those two things sits one of the most demanding computations in nature, and your brain runs it continuously, in real time, without you noticing. It detects edges, tracks motion, separates colors, fuses the two eye images into depth, fills in the blind spot, holds the scene steady while your eyes flick several times a second, and hands you a finished picture in a fraction of a second. Turning impoverished, ambiguous input into an accurate model of the outside world is a reconstruction problem, and reconstruction algorithms of this power are designed, not stumbled upon. A system that computes reality from fragments is the work of a mind.

In full

Vision begins in the retina, itself a piece of neural tissue where photoreceptors transduce light and layers of bipolar, horizontal, amacrine, and ganglion cells already perform edge enhancement and contrast processing before a signal ever leaves the eye. Ganglion axons form the optic nerve, most projecting through the lateral geniculate nucleus of the thalamus to the primary visual cortex (V1) at the back of the brain. From V1 the signal fans out along parallel, specialized pathways: a dorsal stream toward parietal cortex handling motion and spatial location (the where or how pathway) and a ventral stream toward temporal cortex handling form, object identity, and faces (the what pathway). Along the way the brain extracts oriented edges, direction of motion, color, and binocular disparity, the slight difference between the two eyes' images from which it computes stereoscopic depth. It also performs constancy operations, keeping perceived color and size stable despite changing light and distance, and it stabilizes the scene across saccades, the rapid eye movements that occur several times per second, so the world does not appear to jump. The result, delivered in roughly a tenth of a second, is a coherent, upright, three-dimensional percept reconstructed from inverted, flat, incomplete, and constantly moving input. This is a massive, real-time inverse-problem solver. See Specified Complexity and Information Argument for Design.

The mechanism

• Retinal preprocessing. Photoreceptors capture light; retinal circuitry sharpens edges and enhances contrast before the signal leaves the eye.
• Relay to the cortex. The optic nerve carries the signal through the thalamus to primary visual cortex (V1) at the back of the brain.
• Feature extraction. Cortical circuits pull out oriented edges, motion direction, and color in parallel, specialized channels.
• Binocular depth. The brain compares the two slightly different eye images and computes stereoscopic depth from the disparity between them.
• Stabilization and constancy. It fills the blind spot, holds color and size constant under changing conditions, and steadies the scene across several rapid eye movements per second, delivering an upright 3D percept in about a tenth of a second.

Why this points to design

Seeing is not passive reception; it is active reconstruction, and reconstruction from degraded data is precisely the class of problem that requires clever, purpose-built algorithms. The input underdetermines the output: a flat retinal image is consistent with infinitely many three-dimensional scenes, yet the brain reliably recovers the right one by applying built-in assumptions about light, surfaces, and geometry. Solving an underdetermined inverse problem correctly and instantly demands specified information, the equivalent of well-chosen code, embedded in the visual system's architecture. On top of that, the subsystems are interdependent: depth perception needs the two eye images fused and aligned, motion perception needs the scene stabilized against eye movement, and object recognition needs edges and constancy already computed. A pile of unintegrated feature detectors would yield noise, not sight. Integrated, real-time reconstruction that turns fragments into an accurate world model carries the Specified Complexity and functional information that mark design. See Intelligent Design and Information Argument for Design.

The evolutionary account, and why it falls short

The evolutionary account is that vision was refined by degrees because better sight, sharper edges, more reliable depth, faster motion detection, always aided survival, so selection accumulated improvements from light-sensitive spots up to the full mammalian visual system with its parallel processing streams.

Sharper vision is useful, and that is not in question, but the account substitutes the value of the endpoint for an account of the computation. The visual system does not merely have more detectors than a light-sensitive patch; it runs an integrated reconstruction that solves an underdetermined inverse problem, and its parts presuppose one another. Depth from binocular disparity is useless until the two images are precisely registered and the disparity is computed; motion perception is corrupted until the scene is stabilized against the eyes' own several-per-second movements; object recognition needs constancy and edge extraction already in hand. Improving one channel while the others are absent does not yield a better picture but a garbled one, so the graded-improvement story lacks the very intermediates it needs. And the deepest point is untouched: where the built-in assumptions about light and geometry, the priors that let the brain pick the correct scene from infinitely many, came from. Those assumptions are information about the structure of the world, and pointing to the survival value of sight does not explain how that information got into the wiring. That embedded, world-fitting information is what points to design. See Common Descent Critique.

See also

• 50 Amazing Facts About the Human Body, the hub this spoke belongs to
• Intelligent Design, the framework behind the argument
• Specified Complexity, functional information in a reconstruction algorithm
• Information Argument for Design, why solving an inverse problem implies embedded information
• Nerve Signal Speed, a sibling spoke on the fast transmission vision relies on
• The Human Connectome, a sibling spoke on the network that runs the computation