Purchase this article with an account.
Yuri Ostrovsky, Pawan Sinha; Learning to parse images through dynamic experience. Journal of Vision 2006;6(6):674. doi: https://doi.org/10.1167/6.6.674.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Motion has been identified as a primary cue for object segmentation in infant development (Kellman & Spelke, 1983; Johnson & Aslin, 1995). Our work with patients with late visual onset (Ostrovsky & Sinha, 2005) indicates that motion-based segmentation processes are evident soon after sight onset, despite many years of visual deprivation. Because motion analysis is both robust and early, it is possible that motion might serve as a bootstrapping mechanism for the subsequent learning of static segmentation cues such as junctions, line continuity and other figural Gestalt cues (Wertheimer, 1912; Kanizsa, 1979).
To explore this hypothesis, we have developed a class of ambiguous figures using osculating (aka “kissing”) junctions. This junction has two possible interpretations: two curves touching at a point (kissing) or a curved X junction. The kissing interpretation is generally the preferred one. Our experiments suggest that motion is highly effective at disambiguating the figures toward the less-preferred X-junction interpretation, whereas color is relatively ineffective. To investigate a possible role of motion in the learning of object parsing cues, we trained normal adult subjects with these motion stimuli in order to shift the interpretation of the ambiguous junction toward the X interpretation. This shift generalized to other types of figures and even to static presentations, effectively reprogramming the object parsing cues of the visual system. This provides evidence that motion-based segmentation may serve as the base upon which image processing heuristics that are relevant to both dynamic and static displays are learned.
This PDF is available to Subscribers Only