Visual information processing in primate cortex is based on a highly ordered representation of the surrounding world. In addition to the retinotopic mapping of the visual field, systematic variations of the orientation tuning of neurons are described electrophysiologically for the first stages of the visual stream. On the way to understanding the relation of position and orientation representation, in order to give an adequate account of cortical architecture, it will be an essential step to define the minimum spatial requirements for detection of orientation. We addressed the basic question of spatial limits for detecting orientation by comparing computer simulations of simple orientation filters with psychophysical experiments in which the orientation of small lines had to be detected at various positions in the visual field. At sufficiently high contrast levels, the minimum physical length of a line whose orientation can just be resolved is not constant when presented at various eccentricities, but covaries inversely with the cortical magnification factor. A line needs to span less than 0.2 mm on the cortical surface in order to be recognised as oriented, independently of the actual eccentricity at which the stimulus is presented. This seems to indicate that human performance for this task approaches the physical limits, requiring hardly more than approximately three input elements to be activated, in order to detect the orientation of a highly visible line segment. Combined with the estimates for receptive field sizes of orientation-selective filters derived from computer simulations, this experimental result may nourish speculations of how the rather local elementary process underlying orientation detection in the human visual system can be assembled to form much larger receptive fields of the orientation-sensitive neurons known to exist in the primate visual system.
Neural population control via deep image synthesis