Abstract
First-order classification image techniques can reveal the perceptual templates used by an observer to perform a detection or discrimination task. Although the technique has been extended to infer higher-order sub-template features, the typical method of using pixellated noise poses problems of tractability for higher-order analyses. Here we propose a simplified stimulus with the aim of recovering the features involved in letter identification and crowding.
We created a 16-segment display, similar to a digital clock, that can produce letters. In our experiment, one of ten possible target letters was presented per trial in a central display embedded within a larger array of replicated 16-segment grids. I.i.d. Gaussian contrast noise was added to all segments. Observers identified the target and also indicated the position they thought the target appeared. Contrast of the target was controlled by a modified QUEST procedure to maintain performance at 50% correct. We collected 10,000 trials of data for 2 observers in a foveal condition and 2 observers at 10 deg eccentricity.
First-order classification images based on contrast noise were vivid in both fovea and periphery conditions. Observers’ estimations of target position were accurate at the fovea; in the periphery, they were elongated towards fixation. Second-order features defined as noise correlations between unique pairings of segments that could occur within one letter width and height often did not correspond to salient sub-template features of the first-order images. Instead, they revealed an interaction between the noise and signal presented in those error trials. Features used to identify a letter are not necessarily the parts of the letter - they can be the parts of a distractor, which must be perceptually excluded.
Our letter-in-chaff stimulus offers a tractable construction to study the sub-template features involved in letter identification, and potentially uncover the rich feature-feature interactions in peripheral crowding.