Abstract
A fundamental question for vision research concerns how the visual system represents the shapes of three-dimensional (3-D) objects for recognition. To examine this issue we measured eye movement patterns during single object identification. While eye movements have been widely studied in a variety of other domains such as scene analysis, reading, visual search and face perception, there is surprisingly little evidence about eye movement patterns in relation to object recognition. We asked three questions. First, do eye movement patterns show preferences for specific types of image features? Second, are gaze preferences consistent between stimulus encoding and recognition? Third, are they invariant across changes in 3-D viewpoint and illumination? Eye movements were recorded while Ss attempted to first encode and then recognize the shapes of novel 3-D objects each composed of distinct volumetric components. An area of interest (AOI) analysis revealed remarkable consistency in the eye movement patterns between the learning and testing phases, and across different viewpoints of the same objects and lighting conditions. These gaze patterns cannot be predicted from low-level image salience. Rather, the results suggest that fixation patterns during object recognition are driven by higher-level 3-D object representations.