Purchase this article with an account.
Georgin Jacob, SP Arun; We see the forests before the trees but how do they combine? Shape information combines linearly across multiple scales. Journal of Vision 2017;17(10):1247. doi: 10.1167/17.10.1247.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
A classic finding in vision science is that we see the forest before we see the trees. For example, when seeing a circle made of diamonds, humans can report the global shape (circle) faster than they report local shape (diamonds). Although such hierarchical stimuli have been extensively studied, we know little about their underlying representation: how do large and small scale features combine in an object? Do they interact, and if so, how? More generally how does shape information combine across multiple scales? To address these fundamental questions we performed a number of experiments to understand the representation of hierarchical stimuli in perception. In each experiment we measured the perceived dissimilarity between pairs of stimuli using oddball visual search. In Experiment 1, we asked whether the net dissimilarity between shapes differing at both global and local levels could be understood using dissimilarities at the individual scales. Dissimilarity between hierarchical stimuli was explained almost entirely as a linear sum of contributions from shape relations at the global level, at the local levels and from cross-scale interactions across and within objects. Thus, features across multiple spatial scales combine linearly. The scale, position and configuration of local elements can contribute to the percept of the global shape. In Experiment 2, we compared the representation of hierarchical stimuli with shapes made of an external contour and an internal local shape. Both were highly correlated, but we observed weaker cross-scale interactions for exterior-interior shapes. In Experiment 3, these cross-scale interactions increased when the interior shape moved closer to the external contour or increased in size. Taken together, our results show that shape information across multiple scales in an object combine linearly, and delineate the factors that influence cross-scale interactions.
Meeting abstract presented at VSS 2017
This PDF is available to Subscribers Only