Abstract
Glass patterns are random-dot patterns, where a number of dots is paired to dipoles according to a geometric rule (e.g. translation, rotation). Although their local structure is just dot pairs, a global form can be immediately perceived, making them useful stimuli for investigating form perception. A central question has been, whether Glass patterns are detected by global form detectors tuned to specific forms or by more local mechanisms. Pattern density can provide information about local and global processing strategies, as the number of random dots in the pattern affects especially the local structure of the pattern, perceptually deteriorating the salience of individual dipoles. In some previous studies, density has been found to affect the detectability of the pattern, whereas in others, no effect was reported. Here, we wanted to re-examine the issue. Furthermore, we varied the pattern extent in order to investigate both the effect of density and area summation.
Both rotational and translational patterns were used. Dot size was 4.3 and dipole length 8.6 or 25.8 arcmin. Stimulus density was varied from 0.27% to 27% coverage and radius of the circular area from 3 to 6 degrees. Possible edge artefacts were controlled by smoothing the border of the pattern with a Gaussian window. Thresholds for detecting Glass patterns were measured using a 2IFC task, where observers discriminated between a comparison stimulus consisting of coherently oriented signal dipoles (and random-dot noise) and a standard stimulus consisting of dipoles in random orientations. Proportion of signal dipoles was varied.
Results show that threshold vs. density curves are U-shaped, having their minima at middle densities. Area summation was found across the conditions, it being especially prominent in high-density patterns. Interplay between signal density and transformation type was found: maximum detectability is achieved at lower densities for translational patterns than for rotational.