Abstract
Previous research has tended to study the diagnostic information that facilitates scene categorization at a particular level of detail (e.g., the ordinate level: a room, a field, etc.). However, comparatively less research has directly compared the diagnostic information required for categorizing the same scene at multiple levels of specificity. For example, would the diagnostic features that help categorize a scene at the ordinate level as a room also help to categorize the same scene at the subordinate level as a bedroom, or would different information be required? Here, we examined whether the diagnostic information required to categorize scenes at different levels of specificity are located in similar or different spatial regions. In order to address this question, we presented participants with four ordinate-level categories of gray-scale scenes, with each category having exemplars from a further four subcategories. Participants categorized each scene either at an ordinate (e.g., room) or subordinate (e.g., bedroom) level in a four alternative-forced-choice task. Participants viewed scenes through a gaze-contingent window that presented full-resolution information to the fovea and low-passed information to the extrafoveal visual region. Participants therefore always had access to a scene's coarse, spatial-layout, while finer, local details were only available through fixation. By analyzing the resulting fixations as a function of task, we investigated whether the diagnostic information for categorizing scenes at the ordinate and subordinate levels were the same or different. Fixations directed toward similar regions across categorization tasks would indicate that similar diagnostic information is required, while fixations directed toward different regions would suggest that different sets of information are diagnostic. Analyzed behavioral and fixation data showed that in most of the subcategories, the distribution of fixations differed significantly. The results therefore suggest that the diagnostic information required for making categorizing judgements often occur in separate spatial regions of the scene.
GLM was supported by an ESRC Grant (PTA-026-27-2500).