Structural modeling of spatial vision.

A linear structural model of mechanisms underlying spatial vision was generated from the variance-covariance matrix of contrast sensitivity data. The data had been collected on a large group of observers ranging in age from 19 to 87 yr, using gratings of 0.5-16 c/deg spatial frequency. Structural models incorporating various numbers of spatial frequency-tuned mechanisms were compared, with a three-mechanism model giving the best account of the data. The same analysis was applied to contrast sensitivity data simulated from Wilson and Bergen's (1979) model. When the stimulated data covered 0.5-16 c/deg, only three mechanisms were needed to give a satisfactory account of the data; when the simulated data covered 0.25-16 c/deg, four mechanisms were required. Peak sensitivities and bandwidths of the mechanisms extracted from the simulated data resembled those extracted from the real, population data. This reinforces the idea that at an early stage of human vision a small number of spatially-tuned mechanisms are operative. In addition, the outcome of analyses with both real and simulated data illustrate the potential value of linear structural models for vision research.