Dr. Brian Keane
Perceptually representing objects is crucial to thought and action. In this talk, I discuss three lines of research that reveal and relate multiple levels of object representation. In part I, I discuss the filling-in that occurs when object representations are built from spatiotemporally fragmented input. I present data from a recent “classification image” paradigm and make the case that the construction of object representations should be considered a predictive, spatiotemporal process. In part II, I consider how completely disappearing objects can be represented as persisting. I suggest that although the visual system can maintain object represen-tations through impressively long disappearance durations, it does not predict object reappearance on the basis of pre-disappearance trajectories. In Part III, I consider the extent to which featural properties of objects (shape, orientation, etc.) are automatically encoded. By way of a novel paradigm that combines contour interpolation and multiple object tracking, I show that object features are probably richly encoded well before the stage of attentional selection. These investigations suggest important differences between the processes that build and maintain object representations. Object formation processes are predictive and highly sensitive to object features, whereas tracking is generally neither. I conclude by discussing how these results bear on current theories of visual object representation.
| 
