Visual Perception of Contours, Surfaces, and Objects: Basic Research and Modeling
We have a strong and ongoing research effort involving psychophysical research and computational modeling of object perception, including how objects, contours, and surfaces are perceived from information that is fragmentary in space and time, and how we perceive and represent shape.
In recent work, we have developed a neural-style implementation of early contour connections underlying connections of contour fragments across gaps.
In ongoing efforts, we are developing a more comprehensive framework for understanding object formation – extending the basic geometric relations governing interpolation (generation of illusory and occluded contours and surfaces) from 2D static cases to 3D perception and spatiotemporal object perception, in which objects are constructed across gaps in both space and time.
We are applying classification imaging techniques to the problem of spatiotemporal interpolation, exploring the types of motions that support dynamic object formation, and studying object formation across gaps in the context of multiple object tracking. It is hard to see any shape from the three static images. However, when the static images get repeated over time a triangle emerges.
Spatiotemporal Interpolation and Shape Representation
The information we receive from the world is often fragmented in space and time. Objects are often occluded and their parts are only revealed through motion. Nevertheless, we automatically perceive the world as organized into perceptual units and objects: we see a squirrel running through branches, not a bunch of squirrel bits seen through leaves.
How does the visual system integrate visible information and interpolate parts of the scene that are hidden from view? What properties of the scene engender contour and surface interpolation? How are perceptual units formed over time?
We investigate these aspects of shape perception and representation using spatiotemporal boundary formation (SBF) displays (Shipley & Kellman, 1994, 1996). In these displays, clear boundaries and surfaces are formed in the absence of edge information. Change in element properties is required for the perception of form, so the shapes are revealed gradually over time. Take a look at the videos!
Interpolation Automatically Directs Attention in Multiple Object Tracking
Multiple object tracking has almost always been used to explore attention. But we show that a version of the paradigm—which we call Multiple Vertex Tracking—can also be used to explore contour interpolation.
In this paradigm, a target and distractor object orbit in each quadrant of the screen. Four targets can either interpolate with one another to form an illusory quadrilateral (TI condition); or each target can interpolate with two of its neighboring distractor objects (TDI condition). The first condition is consistently easier than the second.
Test it for yourself! Track the blinked objects in the videos: Left: All targets interpolate, so that tracking is easier. Right: Targets and distractors interpolate together, to make tracking harder.
- Philip J. Kellman
- Gennady Erlikhman
- Everett Mettler
- Suzy Carrigan
- Nicholas Baker
- Jennifer Mnookin (UCLA)
- Itiel Dror (UCL)
- Hongjing Lu (UCLA)
- Patrick Garrigan (SJU)
- Brian Keane (Rutgers)
- Tandra Ghose (TU Kaiserslautern)