Abstract
The phase spectra of natural scene imagery play a central role regarding where contours occur, thereby defining the spatial relationship between those features in the formation of image structure. Thus, we were interested in 1) measuring the relative amount of local spatial phase alignment needed by humans to extract contours from an image, and 2) determine if those measurements depended on the contour “sparseness” at different spatial frequencies (SF). We examined this with a match-to-sample task that used either natural scene images or noise images possessing naturalistic contours, grouped with respect to their level of sparseness. Phase alignment in the stimuli was controlled by band-pass filtering the phase spectra, where phase angles falling within the filter's pass-band were preserved, and everything else randomized. Filter widths were varied (0.3 octave steps) about one of three central SFs (3, 6, 12cpd). On any given trial, following a 250ms presentation of a partially phase-randomized image, participants were simultaneously shown (2sec) four images and asked which one corresponded to the previously viewed, partially phase-randomized image. Results indicated that 1) the bandwidth of local spatial phase alignment needed to match image contours depended on the relative sparseness of the original image; 2) for contours falling within the 6cpd central SF filter, less phase alignment was needed as compared to the other central SFs; 3) contour sparseness outside of a given filter's central SF was not found to interfere with the amount of phase alignment needed to match image contours.
Canadian Institutes of Health Research (CIHR) grant: MT 108-18 to RFH