Abstract
Researchers have suggested that responses to multiple visual stimuli can be transformed into a single population response, which serves as a representation that enables people to quickly extract ensemble information from multiple visual elements (e.g., average orientation of tilted bars). This population response account assumes that the visual system disregards location information from multiple visual features and aggregates them as the same population response. The current study explored this concept of a representation built with a bag of free-floating visual features. We created visual stimuli by applying the Fourier transform to either a Gabor patch or a natural image, and then adding noise to either spectral power or phase distributions, independently. Participants were tested with two different tasks, which were associated with different types of stimuli and noise. First, participants judged the average orientation of multiple Gabor patches with random noise added to the spectral power/phase distributions. We found that noise added to the locations of sine-wave components (i.e., random phase noise) did not have much influence on the average orientation judgment. Second, participants categorized natural images with noise added, but this time, noise was derived from other natural images, and thus made systematic biases in the spectral power/phase distributions. If an ensemble comprises free-floating visual features, participants would not be able to filter systematic biases in the location information, and that would disrupt categorization performance. As expected, participants were not good at categorizing natural images with phase noise derived from other, irrelevant natural images. We suggest that an ensemble representation is built from a bag of visual features, and their location information is not bound to each visual feature. Thus, an ensemble is resistant to unbiased random phase noise, yet prone to systematic biases in the spectral phase distribution.
Meeting abstract presented at VSS 2018