Abstract
In the periphery, a letter target can become unrecognizable when flanked by other letters. This “crowding” phenomenon is largely absent in the fovea. Is crowding associated with changes in the features extracted by the visual system? Standard classification-image methods cannot reveal low-level features because it cannot distinguish conjunctive from independent use of features. Even if a clear template is revealed, it does not follow that all parts of the template are utilized as a whole. In this study, we investigated feature utilization by applying a pixel-wise covariance analysis to the noise fields from the error trials of a letter-identification experiment (“o” vs. “x”), run with four conditions: foveal vs. peripheral viewing, with or without flankers. The results were visualized as maps of correlation coefficients between pairs of pixels. In the fovea, the elementary features were found to be curved segments for “o”, and bipolar bar segments for “x”, of roughly the full letter size. The presence of flankers led to stronger inhibition zones without affecting feature shape or size. In the periphery, the features were bar fragments (vertically oriented for “o”, diagonally for “x”), roughly half the letter size. The presence of flankers further reduced the feature size by 1/3, and although the features retained similar shapes, they were surrounded by stronger and tighter inhibition zones. These findings suggest that, compared to the fovea, the features utilized in the periphery are smaller fragments of the target. Their sizes are further reduced by the presence of flankers, which leads to crowding.
Supported by: NIH/NEI R03-EY016391