Supposing that crowding is compulsory grouping suggests a remarkably simple model for object recognition

Abstract

Grouping and crowding have each received much study, but, because they can be avoided, they have seemed marginal to the mystery of how people recognize objects. We begin by showing that the same image parameters that promote grouping also promote crowding. Joining grouping and crowding theories suggests a unified account in which grouping and crowding are two aspects of the classification underlying all object recognition. In this unified account, recognition of a simple object, like a letter, is mediated by a single combining field driving a simple classifier that can only recognize one thing at a time. It follows that the human observer has a vast array of combining fields, one for each size and retinal location, and is usually free to use whichever best matches the target. For simplicity, we suppose they are all similar, differing only in position and size. As eccentricity increases, the smallest available combining field size increases, presumably because fewer neurons are available. Crowding arises when the smallest available combining field is too large to isolate the target object from adjacent objects, as a simple classifier cannot recognize two things at once. To test the notion that the same classifiers underlie both crowding and grouping, we show that the efficiency of identification by a single isolated combining field in the periphery is practically the same as the efficiency of unrestricted central vision of the same target. We conclude that the same unit for object recognition underlies both (optional) grouping and crowding, which is compulsory grouping. Critical spacing marks the boundary between optional and compulsory. Meeting abstract presented at VSS 2015