Abstract
Saliency based image computations have been useful in understanding the mechanisms of guiding overt and covert attentional shifts and predicting eye position in natural and artificial scenes. With this foundation we investigated the relationship between memory, eye position and saliency. Particular we tested the hypothesis that in natural scenes an object's computed saliency will positively correlate with subjects' memory of the object independent of eye location. To this end we recorded eye movements from fourteen naive subjects while they were shown 18 images (subtending 55°x33° of visual angle) from scenes of shopping environments for two second followed by a random mask. Subjects were then asked to recall whether subsequently presented image patches contained items that were present in the scene. We found that the amount of computed saliency of object patches in natural scenes has no significant correlation with subjects' recall rates; however, eye position and fixation time on an object are strong factors in facilitating recall(p [[lt]].05). We also find that saliency can predict eye location three standard deviations above chance. These results indicate that saliency's contribution to memory is mostly through overt(eye movements) and not covert attention.