Abstract
Observers are capable of making rapid and accurate judgments of the summary information in an array composed of multiple elements. Whilst some theories have argued that this suggests that the visual system is capable of automatic extraction of statistical information from a visual scene, others have suggested that serial strategies may play a role. To test whether perceptual averaging occurs in parallel or in series, we asked observers to judge the average feature (shape or colour) in a centrally-presented visual array with a variable number of elements (‘squircles’, i.e. shapes that varied continuously from shape to circle). As reported previously, we found that the variance (heterogeneity) of the array slowed responding and increased error rates, but set size (2, 4 or 8 items) had no influence on performance. Reasoning that it might be possible to average in parallel across information on a single dimension (e.g. shape or colour) but not a conjunction of two dimensions, we devised a new averaging task in which the decision value was continuously signalled by the conjunction of two dimensions, with more red/square or blue/circle items belonging to one category and more blue/square or red/circle items belonging to the other. Surprisingly, for these stimuli we found no influence of the array variability and an inverse set-size effect, with better performance for larger arrays. This latter finding cannot be due to increased precision for larger set sizes (or less variable arrays) because in both experiments elements were pseudosampled to ensure a fixed mean value. Together, these findings suggest that averaging of both one- and two-dimensional information can be conducted in parallel, supporting models suggesting that observers automatically extract summary information from visual scenes.
Meeting abstract presented at VSS 2013