Abstract
Under normal viewing conditions, letters are less visible in the periphery than in the fovea. Letter visibility in the periphery decreases further when a letter is flanked by other letters (“crowding”). Classical ideal-observer analysis attributes changes in performance (accuracy or threshold) to two factors: a change in the equivalent noise internal to the observer and/or a change in the optimality of the underlying computation, measured in terms of sampling efficiency. We extended the ideal-observer analysis to include a limitation in spatial resolution as defined by the contrast sensitivity function (CSF) at the eccentricity of interest. A linear filter, having the shape of the subject's CSF, was placed in front of an ideal observer. This is equivalent to limiting an ideal observer with correlated Gaussian noise instead of white noise. Three observations were made using this CSF-limited ideal-observer model. First, by measuring contrast thresholds for identifying single band-passed filtered letters, we found that for letter identification, spatial tuning in the periphery (5 and 10 deg eccentricity) was the same as that predicted by the CSF-limited ideal-observer model (Chung, Tjan, Legge, 2002, Vis. Res.) and was therefore optimal after discounting the CSF. Second, spatial tuning remained nearly optimal when the target letter was flanked on both sides with filtered letters of the same spatial frequency (peak tuning frequency was only 0.2 octave higher than that exhibited by the model; Chung & Tjan, 2002, ARVO). Third, the substantial threshold elevation observed in letter crowding was due to an increase in the subject's equivalent noise, and not to any reduction in sampling efficiency (Tjan et al., 2004, VSS). In addition, we found that correlated Gaussian noise patches, having the power spectrum of a letter, effectively crowded the target letter, suggesting that the sufficient condition for crowding in the periphery is that the target and its flankers are close in spatial position and spatial frequencies (Tjan & He, 2004, VSS).
Supported by NIH Grant EY12810 (Chung); James H. Zumberge Faculty Research & Innovation Fund and USC Undergraduate Research Grant (Tjan).