Abstract
Being able to categorize our external visual environment is a critical preliminary step for interacting with the world. How we categorize our environment affects subsequent behaviors: e.g., categorizing a scene as a library would suggest proper social etiquette, location of task-relevant objects, etc. Because of the speed with which a scene can be processed, gist processing – that is, what can be discerned within a single glance – has been a frequent topic of research that has promoted a feed-forward processing mechanism. However, categorization can extend beyond gist: scenes can be categorized at different hierarchical levels of specificity; the same scene can be categorized at the superordinate (e.g., indoors), basic (restaurant) and subordinate (diner) level. In our study, we ran two experiments in order to elucidate a top-down component of scene categorization. Participants were asked to categorize scenes at either the basic or subordinate level as a function of low-pass filtering. Basic level categorizations reached an above chance level of performance at a lower-filtering level than subordinate judgments, indicating that diagnostic information changes as a function of the level of category specificity. In a second experiment, scenes were low-pass filtered to a level where the basic level gist of a scene was recognizable but accuracy in determining the subordinate level category was at chance, while a gaze-contingent window showed full resolution information to the fovea. By summing fixations made when making either basic or subordinate judgments, we recorded what high-spatial frequency information participants needed in addition to the low-pass information in order to make accurate judgments. The results showed that the scene regions considered diagnostic changed with the level of category specificity. This suggests that there is a bi-directional interplay between available image features and task-constraints when determining a scene’s category.
Meeting abstract presented at VSS 2012