Abstract
The centroid task requires subjects to use a mouse to indicate the center of a briefly flashed cloud of, typically, dots. Subjects can also judge the centroid of a cloud of highly diverse items that differ in color and shape. Similarly, subjects can judge motion direction in successive frames in which the only thing that changes consistently is an area defined as figure, the substance of both figure and ground changing in each new frame. The fact that subjects can make centroid, motion direction, and other judgments that simultaneously involve the locations of highly different items defined merely as figure versus ground suggests that these computations act on a salience map that records the presence and the location of items but is indifferent to their substance. In motion and centroid tasks, Ss can also selectively respond to attention-selected subsets of items and ignore distracter items. Here we show that, in a single brief flash of either 12 or a 24 dot cloud, with equal numbers of interleaved black, red, and green items, subjects can accurately report the centroids of all three colors. Their performance is further analyzed in terms of three computationally defined components: accuracy of their 3 attention-defined color filters, the fraction of stimulus items incorporated into the centroid computation (efficiency), and mouse misplacement error. Comparing triple response and single-response controls shows: no loss in the quality of 3-versus-1 attention filters, only a 6% overall efficiency loss in judging 3-versus-1 4-dot centroids, and a 20% efficiency loss in 3-versus-1 8-dot centroids. That three centroid computations can occur concurrently with no loss in the quality of the attention filter and only minor losses in overall efficiency requires a reformulation of the single salience map concept: Our subjects exhibit three concurrently active salience maps.