Abstract
Crowding is the failure to recognize a simple target because it is too closely surrounded by clutter. It is characterized by a crowding distance beyond which clutter has no effect. We take this oval area to be the receptive field of a crowding “unit”. Thus, a word seen peripherally can be recognized only if neighboring letters are at least a crowding distance apart, i.e. seen by distinct crowding units. Some objects, e.g. faces, are substantially more complicated than a roman letter and seem to be analyzed more like a word than a letter. At a given eccentricity such objects can only be successfully identified if the object is large enough that crowding units isolate key parts, like the letters in a word or the features of a face. Efficiency for identifying a roman letter in noise is respectable, roughly 15% (Pelli et al., 2006). The efficiency for a word is much worse, reduced by the reciprocal of the number of letters (Pelli & Farell, 2003). We suppose that each crowding unit has a respectable efficiency, but that the visual system combines them inefficiently, requiring the same energy per unit, regardless of the number of units. This predicts that efficiency for a complicated object will be inversely proportional to the number of crowding units required to see it. We tested the prediction by measuring n and efficiency. For n, we measure the threshold size at 10 deg eccentricity, and divide the threshold area by the known crowding area. We measure efficiency from human and ideal threshold contrasts for identification on a white noise background. We tested eight fonts: Hiragino, Songti, Checkers, Kuenstler, Sabbath, San Forgetica, and Sloan. Plotting log efficiency vs. log n, we get a linear regression slope of −0.84 with r=0.82. This supports the hypothesis that recognition of complex objects inefficiently combines the contributions of multiple crowding units.
Acknowledgement: 1R01EY027964