Purchase this article with an account.
Seha Kim, Shaheera Sarwar, Manish Singh, Jacob Feldman; Local and global cues to depth in line drawings. Journal of Vision 2014;14(10):728. doi: https://doi.org/10.1167/14.10.728.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
We studied the influence of local and global information on the 3D interpretation of line drawings of smooth objects. T-junctions are well known to play an important role in conveying relative depth. In previous studies (VSS2013) we found that (1) local depth ordering from T-junctions is probabilistic, with subjects' interpretation of relative depth based on inferred depth difference; and (2) estimates of depth differences propagate spatially from T-junctions along internal contours, suggesting that local and non-local cues are combined to yield a final depth interpretation. Here we examine the interaction of local and global cues in more detail by varying the aperture size through which a line drawing is visible. We created line drawings by projecting randomly-generated 3D shapes (single-axis shapes with varying curvature and cross-section). They were displayed as black contours on white background, with no other depth cues. Apertures of varying size were used to manipulate the amount and type of information available. Pairs of probe dots were placed on opposite sides of a contour, and subjects reported which dot appeared closer. We manipulated the position of the probe, the type of contour segment the probes straddled, and the aperture size. We found substantial effects of aperture size: (1) an overall increase in judgment confidence with increasing aperture size, and (2) complex interactions between the size and type of structural information available within aperture. For example, depth judgments were more reliable when the probes straddled the head of a T-junction compared to its stem and when the head was concave rather than convex. The observed effects are broadly consistent with a probabilistic model of 3D shape interpretation in which local cues to depth order, including T-junctions and contour curvature, are combined probabilistically across the shape to arrive at a final estimate of 3D structure.
Meeting abstract presented at VSS 2014
This PDF is available to Subscribers Only