A two-dimensional spatially inhomogeneous model of the visual system is developed. It is based on the properties of cone spacing in the retina and on the hypothesis of uniform neural interactions (lateral inhibitions). Its quantitative predictions of the detection and discrimination (acuity) of various types of stimuli are studied. The model works well with local stimuli positioned at varied eccentricities as well as with extended stimuli (vertical cosine gratings windowed by two-dimensional windows), but a simple threshold detector was found to be insufficient to describe the increase of contrast sensitivity with the number of cycles of the cosine gratings at high frequencies. It is concluded that, even for one-dimensional stimuli, a two-dimensional approach is necessary and that other parameters such as imprecision of fixation, eye movements, and two-dimensional probability summation must be taken into account before resorting to more complex models.
