Purchase this article with an account.
Benjamin Balas, Pawan Sinha; Learning about objects in motion: Better generalization and sensitivity through temporal association. Journal of Vision 2006;6(6):318. doi: 10.1167/6.6.318.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
When learning to recognize a new object, the visual system must learn to bind together disparate images of that object. Temporal association is a candidate mechanism for achieving this binding. By linking images that are close in time, invariance to observed appearance transformations can be achieved. This hypothesis has derived psychophysical support from studies showing that frames in a motion sequence can be perceptually bound in such a way as to lead to predictable errors in subsequent recognition performance; temporal neighbors are found to be harder to discriminate, even if they come from two different objects.
Here, we demonstrate two distinct effects of temporal association using non-rigid objects. First, in contrast to predictions from previous work, we demonstrate that shape discrimination between images actually improves following temporal association. Second, we measure the extent of image binding over time. Our experimental procedure dispenses with explicit recognition judgments (since these are susceptible to cognitive interference) and instead uses a priming task as an implicit measure of generalization. We find that frames within short temporal windows eventually serve as equally effective priming cues, indicating a linkage of their representations. Taken together, these results suggest that temporal association can simultaneously yield improved discriminability between, as well as an implicit linkage across, frames of a sequence. We demonstrate that this pattern of results can arise from a model of object representation in which temporal association leads to coarse coding across a population of object units.
This PDF is available to Subscribers Only