Promiscuity and Pruning: Investigations of a Two-Stage Theory of Contour Interpolation
- Author(s): Carrigan, Susan Burns
- Advisor(s): Kellman, Philip J
- et al.
One of the most important functions of the human visual system is the construction of representations of whole objects and surfaces. Contour interpolation, the process that connects edges across gaps in the input, is critical to achieving these representations. In this dissertation, I will propose, test, and model a two-stage theory of the contour interpolation process. The first stage of this process interpolates connections indiscriminately across all pairs of edge fragments that meet the geometric relatability constraints defined by Kellman & Shipley (1991). These connections are maintained in an intermediate representation. The second stage of the contour interpolation process receives these promiscuously interpolated contours, and the physically specified edges that gave rise to them, as input. At this stage, a variety of other sources of information in the scene are taken into account to determine the modal or amodal appearance of each interpolated contour as well as its ultimate perceptual strength. The information that impacts an interpolation’s perceptual strength includes cues as to border ownership, luminance contrast polarity, equiluminant color contrast, and the corroboration of interpolated contours across multiple spatial scales. The outputs of this second stage determine the subjective appearance of interpolated contours in a scene. Because the outputs of the first stage exist in an intermediate representation, some contours interpolated in the first stage will not appear in our phenomenological perception. However, because they exist at some level of the perceptual process, they have some measurable impacts on perceptual performance tasks. Chapter 2 introduces a set of experiments that reveal the intermediate outputs of the first stage of the contour interpolation process. Chapter 3 introduces a set of experiments that reveals the effect of a variety of scene cues on the perceptual strengths of interpolated contours, as determined in stage 2. Manipulations of cues as to border ownership, amodal surface spreading (including luminance contrast polarity and equiluminant color contrast differences across inducing fragments), and spatial frequency differences across inducing edges are shown to impact the perception of interpolated contours exiting the second stage. Chapter 4 combines manipulations to scene cues to measure the combined magnitudes of their effects. The results reveal that cue combination in the second stage is well described by an additive model but also show that certain combinations of cues lead to reductions in strength that go beyond additivity. A theory as to what is special about these combinations is discussed in Chapter 5.