Perception of multi-dimensional regularities is driven by salience.

A challenge for the visual system is to detect regularities from multiple dimensions of the environment. Here we examine how regularities in multiple feature dimensions are distinguished from randomness. Participants viewed a matrix containing a structured half and a random half, and judged whether the boundary between the two halves was horizontal or vertical. In Experiments 1 and 2, the cells in the matrix varied independently in the color dimension (red or blue), the shape dimension (circle or square), or both. We found that boundary discrimination accuracy was higher when regularities were present in the color dimension than in the shape dimension, but the accuracy was the same when regularities were present in the color dimension alone or in both dimensions. By adding a third surface dimension (hollow or filled) in Experiments 3 and 4, we found that discrimination accuracy was higher when regularities were present in the surface dimension than in the color dimension, but was the same when regularities were present in the surface dimension alone or in all three dimensions. Moreover, when there were two conflicting boundaries, participants chose the boundary defined by the surface dimension, followed by the color dimension as more visible than the shape dimension (Experiments 5 and 6). Finally, participants were faster at detecting differences in the surface dimension, followed by the color and the shape dimensions (Experiments 7 and 8). These results suggest that perception of regularities in multiple feature dimensions is driven by the presence of regularities in the most salient feature dimension.