Humans would have difficulty navigating through and interacting with the three-dimensional world without the ability to perceive depth. It is well-known that humans segment and analyze objects in the visible world using (in part) low-level salient features, such as contrast, color and brightness, but how does depth (e.g., retinal disparity) contribute to saliency? We analyzed the effect of stereopsis on human visual saliency by investigating how human fixations change with and without disparity depth cues. A mirror stereoscope was used to present stimuli with or without a stereoscopic manipulation to subjects while subjects eye movements were recorded. Spatial image features as well as disparity features were investigated both for the locations of the measured fixations, as well as randomly-sampled positions. Eye movements were recorded while subjects viewed two sets of stimuli: one that includes only synthetic objects and ground-truth disparity, and one that includes only natural scenes with a less precisely-estimated disparity. Results differed between the two stimulus sets. For natural scenes, subjects tended to fixate locations with lower disparity than at random locations. However, for the synthetic database, subjects tended to fixate regions of higher disparity than at random locations. These findings were confirmed using a center-surround measure with further statistical analysis over the two stimulus sets.

Results p1

Results p2

Results p3