Eye Movements during Free Viewing Maximize Scene Understanding

Abstract

Eye movements when observers are presented with a scene without a specific task (free viewing) have been explained in terms of guidance by saliency and, more recently, by a region's meaningfulness. We show that free-viewing eye movements are more frequently directed to objects critical to global scene understanding rather than salient or locally meaningful regions. The free-viewing fixations are similar to those of observers instructed to describe scenes. Small visual alterations that change a scene’s meaning, but not the most salient or locally meaningful object, alter where observers fixate. By having observers describe the scene while maintaining fixation on different objects, we show that fixations during free viewing are functionally important to extract the scene's meaning. Thus, we conclude that during free viewing of scenes, observers seek to extract scene meaning and direct frequent eye movements to objects that maximize scene understanding.