Abstract
When similar visual objects interfere in the periphery, the percept is "crowded," and observers have difficulty identifying a target object. We aim to characterize the types of errors produced in crowding by parameterizing subjects’ responses via statistical models embodying principles of spatial substitution and spatial pooling of crowded objects. Subjects saw an array of nine random letters arranged in a semi-circle centered on fixation and reported the identity of the cued target letter. In separate experiments, we manipulated the spacing between items, eccentricity of the letter array and duration of the precue, indicating the target location prior to letter array onset. Because all presented items were unique, we could identify the position of reported intrusions, enabling us to characterize patterns in the subjects’ errors and compare them to errors predicted by models embodying spatial substitution and pooling mechanisms. We find that characterizing subjects’ responses requires accounting for the pairwise confusability of individual letters, as well as variations in flanking letters’ influence as a function of space. However, once both letter confusion and spatial weighting are considered, spatial substitution models (that do not consider non-additive interactions of adjacent letters) and spatial pooling models (that have non-additive interactions of adjacent letters) can capture participants’ performance equally well. Finally, we investigate whether the three experimental manipulations of crowding difficulty (spacing, eccentricity, and precue) had qualitatively different effects on the types of errors that subjects produced. We considered how the parametric distribution of errors changed as a function of condition difficulty and found that all three manipulations had substantial effects on errors, but the error distributions fell on a single dimension of difficulty. Thus, it does not appear that (i.e.) precueing changes the spatial precision while spacing or eccentricity changes the individual letter confusability: all the difficulty manipulations that we tested modulate performance similarly.
Meeting abstract presented at VSS 2013