When participants perform a selective centroid task, they are instructed to attend to a specific stimulus type, and therefore are assumed to deploy top-down attention in order to find the target items in the display. Bottom-up mechanisms that may be characterized as grouping may also be influencing performance. In this thesis, we explore the role that this bottom-up mechanism may have for our ability to compute statistical summaries of a cluster of target stimuli. In the first study, targets and distractors were either homogeneous or heterogeneous. Results showed that homogeneity in both target and distractor groups improves selectivity for target stimuli and that the effect of target homogeneity was greater than the effect of distractor homogeneity. This strong effect of target homogeneity is further supported by the second study, in which participants are not told which stimuli would be the target, but that they needed to find the centroid of the “most numerous color.” To explore whether top-down and bottom-up attention are separate mechanisms or whether they compete for resources within a single mechanism, participants were told to attend either to a specified color or to the most numerous color. One condition of interest in this experiment asked participants to attend to a specific color that was not the most numerous, which caused top-down and bottom-up attention to conflict with one another. Results indicated that participants were able to ignore the most numerous color (which drove bottom-up attention) and attend to the instructed target color (which drove top-down attention). However, salient features prevailed in the third study, in which participants were asked to attend to the color and ignore the size of the stimuli. They were unable to completely disregard the size of the stimuli when making their centroid judgement, which suggests that bottom-up cues may not be obligatory, but they do influence which stimuli within the target group is given more weight. A fourth study was conducted to find when, in relation to the grouping process, luminance constancy occurred. Results hinted that constancy may be determined before grouping occurs but were overall inconclusive.

When presented with a complex scene our visual system relies on perceptions that occur pre-attentively, or in just a fraction of a second. This ability, known as summary statistical representations, allows us to get the gist of a group of items. The studies presented in this thesis are focused on two of these summary statistical representations, the centroid and the mean-feature estimations, and are designed to understand mean-feature estimations in the centroid paradigm framework. First, we compared observers’ performances between mean-size and centroid estimations. There were two different centroid tasks: in one, observers were asked to give more weight to larger items and in the other they were asked to ignore size. In a third task, observers judged the mean size of the same stimuli used in the centroid tasks. Results from this study showed that observers were able to either ignore or use size to make centroid responses efficiently, but that mean-size judgments were inefficient. To asses the generality of these results, in a second study we asked observers to estimate the mean luminance of a group of items and compared their performance when asked to estimate the centroid giving more weight to lighter items and less weight to darker items or to estimate the centroid ignoring the luminance of the items. Again, we found high Efficiency in both centroid tasks, but low Efficiency in the mean-luminance task. Finally, we explored whether the distinction between the ventral-dorsal streams of visual processing as described by Goodale & Milner (1992) might provide a plausible explanation for the difference in performances between the mean-size and centroid task. We found that when the relative mean size of one of three groups was used to select and guide movements (dorsal processing) to that group, observers were again able to use size information efficiently.