Abstract
Everyday, people flexibly perform different categorizations of common faces, objects and scenes. Intuition and scattered evidence suggest that these categorizations require the use of different visual information from the input. However, there is no unifying method, based on human categorization performance, which isolates the information used. To this end, we developed Bubbles, a general technique that can assign the credit of a human categorization performance to specific visual information. Bubbles starts with specifying an ‘image generation space’ (e.g., the 2D image plane, or the 3D space of 2D image locations x n spatial scales). Stimuli are designed to randomly and sparsely sample this space with “bubbles” of information of Gaussian shape. Subjects are instructed to recognize these sparse stimuli. The number of bubbles is adjusted to maintain performance at a set criteria. Subjects succeed when the bubbles reveal enough information; the locations of these bubbles are recorded in a CorrectSpace and a TotalSpace. When subjects cannot identify the sparse stimuli, the bubbles are not sufficiently informative, and we only add their locations to the TotalSpace. We then divide CorrectSpace by TotalSpace to derive a ProportionSpace that weighs the significance of each region of the image generation space for the task at hand. We applied Bubbles to human and ideal observers resolving different tasks of face recognition (identity, gender, expression) and object recognition (basic vs. subordinate categorizations) to isolate the specificity of human feature extraction.