Abstract
Object recognition involves the detection and integration of valid features to form a coherent percept. Previous studies hinted that different parts of letters are used for identification in the fovea and the periphery; and that more valid features in the stimulus are required for identifying crowded than for single letters. However, the specific features used for identifying individual letters are unknown. Are these features different between the fovea and periphery, and between crowded and uncrowded conditions? To answer these questions, we used a set of 26 lowercase letters that were constructed of Gaussian patches, with each patch having the chance to be individually turned on or off. We tracked observers' response (letter identity) to each stimulus letter presented singly, or when flanked by two nearby letters, at the fovea and 10° lower field. Letter size was twice the acuity-threshold separately determined for each condition. On each trial, a random subset of the patches that constituted the target and its flanking letters (if present) was shown. The probability of a patch being shown was adjusted such that each observer's overall performance accuracy was close to 52% correct. We performed a reverse-correlation analysis to identify the patch locations that significantly correlated with each observer's response for a given letter, resulting in a “feature map” for identifying that letter. Contrary to the expectations based on previous studies, the feature maps are very similar between the fovea and 10° lower field, and between the crowded and single-letter conditions. For some letters, there are subtle differences across conditions in how much observers rely on a given patch location. Our findings suggest that when letter size is scaled to equate for overall performance, the critical features for letter identification are largely invariant between the fovea and periphery, and between the crowded and uncrowded conditions.
Supported by NIH grants EY12810 (SC) and EY16391 (BST).