Visual search: Saccade parameters depend on the shape of the search area

Search performance can be improved by restricting search to regions relevant to the task: no time is lost then to search in areas that certainly do not contain targets. Cognitive or physical borders can define relevant regions. We examined how the shape of search areas affected saccadic amplitudes and directions. We recorded eye movements during search in areas having one of three shapes: circular, squared or triangular. The areas were presented in a randomly mixed design. Subjects were instructed to find an unknown number of targets. Search areas were uniformly gray shapes and targets were slightly brighter spots. The results show that mean saccade amplitudes appeared to scale with the largest cross-section of the area. Saccade directions were mainly parallel to the borders of the search area. We constructed a simple model to describe the results. The model assumes a circular visual span of constant size around the fixation location. Within the visual span targets are detected. New fixation locations have to meet two conditions: 1) they must be inside the search area, 2) the visual span should not overlap previously searched areas by more than a certain percentage. The new fixation location is selected randomly from all possible remaining locations. The process is stopped when the visual spans cover 90% of the total area or when there are no fixation locations anymore. Saccades follow from the model as vectors between two successive fixation locations. The model has two parameters: 1) the size of the visual span, 2) the percentage of overlap. Saccade directions from the model showed distributions that were similar to those of recorded directions, although the latter distributions were always more peaked. In conclusion, we found that the shape of the search area strongly influenced saccade generation. The effects were partially described by a simple model whose major components were a constant visual span and an imperfect memory.