Abstract
Crowding refers to the marked inability to identify a target when it is flanked by other objects. Recent findings have linked crowding to many form-vision deficiencies in the periphery (Levi, 2008; Pelli and Tillman, 2008). Currently, no precisely specified computational model exists to satisfactorily explain the phenomenon. We used an ideal-observer model to examine the roles of three types of limitations that might lead to crowding in peripheral vision: low spatial resolution, large positional uncertainty, and a perceptual template with a limited field of view (FOV). Each combination of these limitations was examined. Given the assumed limitations, the ideal observer made the statistically optimal decision to identify a target letter (“a”, “e”, “o”, “s”) either presented alone or flanked by a letter on each side. We gauged the performance of the ideal-observer model by the stimulus signal-to-noise ratio (SNR) required to reach 75% correct identification. Low spatial resolution was simulated by low-pass filtering the stimulus with 1/e cut-off at 2.5 c/letter. Positional uncertainty, if present, was set to +/− 0.5 letter width. FOV was either one letter width or unlimited. These conditions were chosen to mimic the condition of viewing a letter 2∓3 times the acuity at 5° eccentricity. We found that while adding one or more of these limitations elevated SNR threshold, thresholds for the flanked-letter condition were essentially identical to those for the single-letter condition except when all three limitations were present. In other words, crowding occurs for an observer that optimally utilizes stimulus information only when spatial resolution is low, positional uncertainty is high, and the field of view of the perceptual template covers only one object. These requirements explain the weak crowding in the fovea and why the internal features of an object cause less crowding than adjacent objects.
Support: NIH grant R01-EY017707.