Trans-saccadic integration for object recognition peters out with pre-saccadic object eccentricity as target-directed saccades become more saliency-driven

Abstract

Bringing objects from peripheral locations to fovea via saccades facilitates their recognition. Human observers integrate pre- and post-saccadic information for recognition. This integration has only been investigated using instructed saccades to prescribed locations. Typically, the target has a fixed pre-saccadic location in an uncluttered scene and is viewed by a pre-determined post-saccadic duration. Consequently, whether trans-saccadic integration is limited or absent when the pre-saccadic target eccentricity is too large in cluttered scenes in unknown. Our study revealed this limit during visual exploration, when observers decided themselves when and to where to make their saccades. We asked thirty observers (400 trials each) to find and report as quickly as possible a target amongst 404 non-targets in an image spanning $57.3°\times 33.8°$ in visual angle. We measured the target’s pre-saccadic eccentricity $e$, the duration $T_{pre}$ of the fixation before the saccade, and the post-saccadic foveal viewing duration $T_{post}$. This $T_{post}$ increased with $e$ before starting to saturate around eccentricity $e_p=10°-20°$. Meanwhile, $T_{pre}$ increased much more slowly with $e$ and started decreasing before $e_p$. These observations imply the following at sufficiently large pre-saccadic eccentricities: the trans-saccadic integration ceases, target recognition relies exclusively on post-saccadic foveal vision, decision to saccade to the target relies exclusively on target saliency rather than identification. These implications should be applicable to general behavior, although $e_p$ should depend on object and scene properties. They are consistent with the Central-peripheral Dichotomy that central and peripheral vision are specialized for seeing and looking, respectively.