Action potential: also known as a nerve impulse or spike.

Adaptation: see plasticity.

Age-related macular degeneration (AMD): a vision disease that strikes mainly the elderly, its most usual effect being the destruction of cells in the central area of the retina.

Amacrine cells: two types of retinal cells that perform a broad field-erase function. The narrow-field amacrine cells have a narrow spread of processes and primarily feed back to bipolar cells to cut short their activity in time. The widely ramifying amacrines broadcast activity related to change and movement over wide spaces on the retina.

Artificial vision: electronic circuitry for processing and interpreting visual information, presenting the result to the individual through a sensory modality other than sight.

Axon: output fiber from a nerve cell. Each retinal ganglion cell has a single axon, typically about 1 µm wide and 5 cm long.

Bipolar cells: in the retina, the third sheet of cells, which accept the information formed by the interactions of the horizontal cells and photoreceptors and convey it to the inner retina, where further interactions occur. There are two, complementary types: ON bipolar cells respond to increases in intensity; OFF bipolar cells respond to decreases in intensity.

Block matching: a popular machine vision algorithm for computing optical flow. At a particular point (pixel) in the image, a block of, say, 8-by-8 pixels is chosen from the current frame and compared with all 8-by-8 blocks near the pixel in the previous frame. The best match for the subframe yields an estimate of the point's movement.

Convolution: in the case of vision, a linear, shift-invariant operation by which an image is filtered by a particular function.

Dendrites: the regions of the neuron that receive synaptic input.

Early vision: processes that recover the physical properties of the visible surfaces of solid objects from their two-dimensional intensity arrays.

Enhanced vision: assistive devices that present visual information to whatever remains of an individual's sighted retina after processing the image for maximum visibility.

Fovea: the small central portion of the retina, corresponding to about 1 degree of visual field, where a very high receptor density provides the finest resolution for observation of fine spatial detail.

Ganglion cells: the output cells of the retina. Their dendrites reach up into the inner retina and read out activity formed by interactions of bipolar and amacrine cells. The axons of the ganglion cells form the optic nerve.

Gyrus: a bump or bulge lying between the sulci of the brain.

Horizontal cells: the network of interconnected neurons that form a sheet just beneath the photoreceptors. The horizontal cell network is responsible for averaging visual activity over space and time and for controlling the gain and offset of the photoreceptor signal.

Hot mirror: a mirror that reflects infrared light but is transparent to visible light.

Line width (minimum feature size): the smallest line width used in a given semiconductor design. While most university-designed analog chips use 2.0- or 1.2-µm design rules for convenience, high-density memory chips can be built with 0.35-µm technology.

Macula: a larger central region of the retina, including the fovea at its center, that corresponds to about 30 degrees of visual field.

Narrow-field cell: see amacrine cells.

Neural activity: activity in neurons, expressed and measured in millivolts, that is a measure of signal intensity. Graded-potential cells, like most of those in the retina, polarize in the range from about -30 to -50 mV. In spiking cells (like the ganglions), spike frequency is the measure of intensity. A typical neuron is at rest near -70 mV and action potentials polarize neurons to near +10 mV for about 1 ms. Typical sensory system spike rates are up to about 50-60 per second.

Optic nerve: a bundle of about one million nerve fibers that carries the visual message from the eyeball to higher visual centers in the brain.

Optical flow field: the output of a vision algorithm or circuit that assigns every location in an image a vector indicating the direction in which the image intensities moved during the preceding interval.

Parameter space: a set of coordinated parameters describing an object. From the computer graphics perspective, these are the "control knobs" for rendering an image of the object. From the computer vision perspective, the parameters are estimated from an image of the object.

Percept: loosely, the smallest component of visual sensation.

Pericranium: The deepest of the 5 layers of the scalp, closest to the skull.

Phosphene: the sensation of light produced by electrical or mechanical stimulation of the visual pathway of the nervous system.

Photoreceptors: two types of photosensitive cells that convert light to neuronal signals. Rods are the more sensitive and are used in moonlight, say. Cones are less sensitive, operate over a broad range of intensities, and convey color information.

Plasticity: changeability in the performance of a neuronal system, usually as the result of experience and usually manifest as changes in synaptic efficacy. Long-term plasticity may involve physical changes such as the formation of new synapses. Short-term plasticity may involve the enhancement of existing synaptic connections.

Prosthetic vision: processed visual information sent to the inner retina or other visual pathways through electrical stimulation of the surviving neurons.

Retina: the neuronal image-processing system that lines the back of the eye. Most processing done here is performed by spikeless, analog neuronal interactions.

Retinitis pigmentosa (RP): a disease of the photoreceptors, frequently inherited, which at its most advanced stage allows only a tiny island of central vision.

Retinotopy: the notion that receptor cells in the retina are mapped to points on the surface of the visual cortex of the brain.

Synapse: the functional contact between two neurons. Signal propagation can occur by physical contact with a bidirectional flow of ions, as in the ganglions and most neurons, or by chemicals known as neurotransmitters.

Spike: also known as a nerve impulse or action potential, the means by which excitation is propagated along an axon.

Sulci: grooves in the surface of the brain.

Supervised learning: a way of setting the weights of coefficients in a neural network (weights are roughly analogous to the strength of synaptic junctions in biological organisms). In machine learning, it usually indicates training a neural network through a set of examples consisting of a "correct" input and output.

Tuning: essentially, the sensitizing of neurons to specific optimal stimuli, such as a line of a certain orientation moving at a certain speed. A unit (biological or computational) is tuned when its response to specific input stimuli is at its maximum.

Widely ramifying cell: see amacrine cells.

(c) Copyright 1996, The Institute of Electrical and Electronics Engineers, Inc.