| Hi-index | 0.98 |
Visual scanning is used by mammals to integrate information from a visual space larger than that which can be placed simultaneously on the fovea. We describe here mechanisms, emulating the biological, that generate scan paths in an artificial neural system. Reorienting reflexes (analogous to saccadic eye movements) result in a sequence of images, initially ordered by environmental characteristics, that determine the features available for pattern analysis. These sequences are learned through a mechanism of heterosynaptic facilitation (achieving classical conditioning) so that a temporal pattern of perceived features predicts the next spatial pattern that should result from a given saccade. The saccade is induced by shared influences of the previous detected features and the current visual input. Detection of the expected features accelerates the decision to produce additional saccades. A scan path develops as the process continues in time.