Abstract
Peripheral vision is impoverished relative to foveal vision in several ways, including acuity, contrast sensitivity, and color sensitivity. Besides these factors, visual crowding also severely limits peripheral visual recognition: Items within "pooling regions" that grow larger with increasing eccentricity are subject to some form of aggregate processing that tends to reduce observers' ability to correctly identify targets. Besides failures of visual recognition, it has recently been shown that perceived numerosity in the periphery also appears to be affected by crowding. Specifically, numerosity is reduced in peripheral vision, suggesting that the integration process that combines information within pooling regions does so in a manner that systematically underestimates how many items are in an array. Presently, we examined mechanisms of visual crowding that may explain this phenomenon. We have previously demonstrated that summary-statistic representations account for many features of visual crowding and other tasks involving peripheral vision. We therefore applied a summary-statistic model of peripheral vision to object arrays to determine if numerosity was systematically reduced in "mongrel" images that reflect appearance constraints imposed by texture features. For displays of varying size (40, 60, or 80 dots) we found that numerosity was indeed reduced in mongrel images over a range of model parameter values in good agreement with prior behavioral results. We also identified parameter ranges where the model does not match behavior, placing key constraints on the tuning of the underlying model. Finally, by varying the shape and size of the elements in our arrays, we found that our model predicts that peripheral numerosity may be underestimated by varying amounts as a function of these properties, and in some cases may even be overestimated. These are novel, and to our knowledge, untested predictions regarding how number may be perceived in the visual periphery using summary statistics.
Meeting abstract presented at VSS 2016