Abstract
As shown in the classic Dalmatian dog demonstration, top-down knowledge aids figure-ground segregation. Such top-down knowledge is acquired through perceptual learning, and this study introduces a new paradigm to measure how learning affects figure-ground segregation. Subjects were required to detect symmetric dot patterns (targets) embedded in a random dot background field. The targets appeared on either the left or right side of fixation at a fixed eccentricity, and subjects reported the location of the target. Each subject was trained on two different dot patterns. To facilitate training at the beginning, bottom-up segregation cues were provided; targets were colored yellow amongst a background field of white dots. As learning proceeded across blocks, the color of the targets gradually changed to white, so that by the end of training, the dot field appeared homogeneous. During the testing phase, subjects were tested on both trained patterns and novel patterns using the homogenous displays. Subjects were significantly better at detecting trained patterns. This result allows us to study how the visual system acquires top-down visual knowledge to facilitate figure-ground segregation.