Abstract
Substantial evidence suggests that observers can accurately estimate the centroid of a spatially extended target. We investigated the top-down attentional control of these mechanisms. Observers estimated (with mouse-clicks) the centroids of briefly flashed, sparse clouds of either 8 or 16 dots of various intensities under three different attentional instructions: give equal weight (i) to just those dots brighter than the background, (ii) to just those dots darker than the background, and (iii) to all dots. Under all conditions participants did well at achieving the required attentional filter. We then required observers to repeat centroid estimations based on the same three attentional instructions, but to weight pertinent dots in proportion to their contrast amplitudes, assigning more weight to dots with extreme contrasts. Results: Observers were able to impose slightly different intensity-selective filters for Proportionally Weighted centroids compared to Equally Weighted ones. In both the Equal Weighting and Proportional Weighting conditions, a decrease in attentional efficiency was observed as target size was increased from 8 to 16 dots. A separate analysis of the centroid computation itself showed high variability across participants in the relative weights attributed to centrally versus peripherally located dots. Some observers down-weighted the peripheral dots relative to more central dots, other did the opposite. Our model-based analysis of centroid judgments yields a quantitative description of the multiple attention filters that subjects use to select certain dot intensities or centroid processing and of the subjects' distance-weighting functions used to compute the centroid.