Datensatz

Zusammenfassung

Patterns of abrupt changes in a scene, such as the dynamic occlusion of texture elements (causing
their apppearance and disappearance), can give rise to the perception of the edges of the
occluder via a process called Spatiotemporal Boundary Formation (SBF). It has previously
been shown that SBF can be disrupted by very small amounts of dynamic noise spread globally
throughout a scene. We recently developed a mathematical model of SBF in which groups
of local changes are used to extract edges, which are then combined into a gure and used
to estimate the gure's motion. The model implies that SBF relies on local processing and
predicts that SBF should be impaired when noise is added near the edges of the gure, but
not when it is added far from the edges. This prediction was tested in a shape-identication
task in which the location of noise is varied. Indeed, performance was not impaired by noise
far from the gure, but was markedly disrupted by noise near the gure, supporting the notion
that changes are integrated locally rather than globally during SBF. In the second part of this
project, the mathematical model of SBF was implemented in software. Reichardt-based motion
detectors were used to lter the experimental stimuli and provide the input to the software
implementation. Three simple geometrical gures, similar to those used in the psychophysical
experiment, were reconstructed using this method, demonstrating one way in which a mid-level
visual mechanism such as SBF could connect low-level mechanisms such as change detection
to higher-level mechanisms such as shape detection.