The Rapid Segmentation of Multiple Objects Is Based On Global Rather Than Local Sampling
We previously showed that people can discriminate multiple intermixed groups of objects based on "segmentability," large gaps between values in feature distributions forming several peaks. Here, we test whether such discrimination is based on local or global sampling. Two arrays of lines of various orientation (O) and length (L) were presented; both had identical feature distributions but opposite directions of O-L correlations. These sets consisted of either 14 lines near both meridians or 32 lines filling rectangular regions; participants had to determine boundary orientation between the sets with different O-L correlations. We found that displays with both O and L segmentable provide better discrimination than nonsegmentable ones and an advantage of 32-line sets. This suggests that the segmentation of spatially mixed objects is global sampling of lots of items based on full-scale feature statistics rather than local sampling near a potential boundary.