Abstract
Humans can rapidly and accurately estimate small number of objects without relying on counting, a process known as subitizing. Objects in natural scenes rarely occur in isolation but are surrounded by other objects or embedded in textured surfaces. Here we investigate how spatial organisation of context elements affects numerosity estimation and whether numerosity mechanisms are selective to elements’ colour, luminance polarity and orientation. Stimuli consisted of a small number (3 to 6) of target elements (Gaussian blobs) presented either in isolation or embedded in context elements. The target and context elements were dissociated by luminance polarity. We varied spatial configuration of (i) target elements by placing the elements either mirror-symmetric, on the vertices of simple geometric shapes or random, and (ii) context elements, by organising the surrounding elements either in a grid, mirror-symmetric, translation-symmetric or random. To examine the selectivity to target-elements’ features e.g., colour, luminance polarity, orientation (Gabors), we compared target-only conditions in which all elements were either the same or different in one of these particular features. We measured accuracy and reaction times (RTs) while participants performed a numerosity estimation task. We found (a) lower accuracy and slower RTs for all target-types when presented in context than in isolation, and for larger (5,6) than smaller (3,4) target numerosity; (b) significantly better performance with the grid compared to mirror-symmetric, translation and random contexts, except for 6 target-elements condition where the grid context yielded the lowest accuracy; (c) comparable performance in the same and different feature conditions for colour, luminance polarity and orientation in the absence of context; (d) better performance with shape than symmetric and random positioned target-elements. We conclude that numerosity mechanisms are not selective to colour, luminance polarity and orientation, and that symmetric, translation and random organisations of context elements inhibit target-numerosity encoding stronger than regular/grid context.