The purpose of eye movements.

the sine-wave spatial frequency and orientation tuning of the cells and the two-dimensional Fourier spectra of the stimuli, since the cells fire to a stimulus to the extent to which the Fourier spectrum of the stimulus has energy within the spatial frequency and orientation range to which the cell is sensitive. There has been a traditional long-standing emphasis in vision on edges and contours. The usual experimental and clinical tests of spatial vision, for instance, involve determination of the finest detail which a person can resolve. Such edges, contours, and fine details are produced by high spatial frequencies. This bias toward considering only high spatial frequencies is partly due to historical reasons, partly to theoretical emphases on edge detectors, and partly to the fact that a clinical loss of high-spatial-frequency vision is so common and is usually simply correctable with glasses. There are many arguments one could offer against such an overemphasis on high spatial frequencies. Among them are the fact that the vast majority of cortical cells are responsive only to low or middle spatial frequencies and that although high-spatial-frequency information may be important in perceiving the fine structure within objects, it is principally low-spatial-frequency information which defines the objects themselves. These points are of importance in their own right, but they gain relevance here from the fact that color information is mainly present in the visual system at lower spatial frequencies and plays a far more important role in spatial vision than has traditionally been credited to it. The cones containing different photopigments should not be thought of as color receptors, but just as light receptors. The long (L) and middle (M) wavelength cones, at least, respond to light across the whole spectrum. These cones, then, respond to visual objects of any orientation, size, shape, or color, whether stationary or moving. Information about each of these attributes is extracted by later neural elements, in each case by comparing the outputs of different receptors. Luminance variations in the retinal image (variations in the number of photons captured) drive the different receptor types in synchrony, the receptors all depolarizing to decrements and hyperpolarizing to increments in light. Pure color changes (variation in the wavelength but not in the number of photons captured), however, produce opposite responses from the different cone types, an equal-luminance shift toward long wavelengths depolarizing the M cones and hyperpolarizing the L cones, and vice versa for a shift toward short wavelengths. The vast majority of ganglion and geniculate cells (in monkey and presumably in humans) receive opposite inputs from two cone types, and these inputs are also spatially separated, e.g., M cones excitatory in the RF center and L cones inhibitor)' in the RF surround. In response to a luminance change (which drives the receptors in the same direction), then, such a cell will show center-surround antagonism and have a small RF center. As a consequence, the cell will be responsive to high spatial frequencies and show attenuation in sensitivity to low spatial frequencies. In response to a color change (to which the receptors respond in opposite directions), however, the same cell will show center-surround synergy, not antagonism, so that the optimum color stimulus is a full-field color change. The cells, in other words, are insensitive to high spatial frequencies but show no low-spatial-frequency attenuation to pure color variations. The information arriving at the cortex about luminance variations on the retina is thus mainly at middle and high spatial frequencies, but about color variations it is at low and middle spatial frequencies. It appears that color and luminance information is processed veiy similarly by cells in the striate cortex. We find cortical cells to have a powerful center-surround organization (and thus narrow bandpass spatial tuning) and orientation selectivity in the color domain as well as in the luminance domain, as discussed above. An array of cortical cells thus appears to dissect the incoming color information from a patch of retina into its twodimensional Fourier spectrum (spatial frequency at a given orientation) as well. In conclusion, it is clear that both color and luminance variations play important roles in spatial vision. Color variations are analyzed mainly over the low-to-middle spatial-frequency range and play an important role in object perception. Luminance variations are analyzed mainly over the middle-to-high spatial-frequency range and play a role in object perception and also in the perception of fine detail within objects.