Abstract
Objects that are semantically related to their scene context are better recognized than objects shown in unrelated surroundings. While this object-scene consistency effect has been reported many times, it remains unclear which information of a scene is used to facilitate semantic object processing. Here, we specifically investigated whether coarse global scene properties preserved by low spatial frequencies (SFs) – that have been shown to mediate a scene's "gist" – drive the consistency effect. In two experiments, we manipulated semantic object-scene consistency by presenting indoor or outdoor objects superimposed on indoor or outdoor scenes. On each trial, a consistent or inconsistent object-scene pair was displayed for 50 ms followed by a dynamic mask. Observers' task was to name the foreground object as precisely as possible, and afterwards report how confident they were about their answer. In Experiment 1, the background scenes were either unaltered, low-pass filtered at 8 cycles per image (cpi) – corresponding to 0.31 cycles per degree of visual angle (cpd) – or scrambled by randomly re-arranging all pixels. In Experiment 2, the scenes were either presented unaltered, low-pass filtered with a higher cutoff (17 cpi; 0.66 cpd) or as hybrid images, selectively preserving the SFs found to be diagnostic for rapid basic-level scene categorization (Willenbockel, Gosselin, & Võ, VSS 2017). In both experiments, in the unaltered scene condition, participants named consistent objects more accurately than inconsistent objects, whereas no such consistency effect was found for any other condition. The confidence ratings were unaffected by the consistency manipulation across conditions. Thus, while low SFs mediate scene gist, we have found no indication that these low SFs have behavioral effects on object processing. Recording event-related potentials might provide an alternative, possibly more sensitive, approach to assess the contribution of coarse global scene properties to the object-scene consistency effect.
Meeting abstract presented at VSS 2018