Purchase this article with an account.
Philip Kellman, Gennady Erlikhman; Understanding and Modeling Spatiotemporal Boundary Formation. Journal of Vision 2015;15(12):525. doi: https://doi.org/10.1167/15.12.525.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Background: In spatiotemporal boundary formation (SBF), illusory contours, shape, and global motion are seen from sequential transformations of sparse texture elements (Shipley & Kellman, 1994). Formally, local edge orientation can be recovered from small numbers of pairs of element transformations; these “motion signals” are not seen as local motions but are integrated to determine edge orientation. An ideal observer model that uses pairs of transformations outperforms human observers. Evidence suggests information integration in SBF is limited to a 165 ms temporal windows and asymptotes at about 36 element changes. In addition, noise in encoding of distances, angular relations, and intervals between transformations may also affect performance. We measured these noise sources experimentally, incorporated them into the model, and tested the model. Design: Displays consisted of black dots on a white background. An invisible bar moved across the display. Whenever the bar contacted a dot, it disappeared, reappearing when the bar passed. Subjects judged the bar as tilted clockwise or counter-clockwise. Discrimination thresholds were measured in three SBF experiments as a function of element density, number, and frame duration. Three non-SBF, 2IFC experiments tested discrimination of distances, angular relations, and temporal intervals of element transformation pairs. Results: SBF performance improved with increasing density and number, and worsened with increasing frame duration. Noise from psychometric functions fit to human data for precision in discriminating distances, angles and intervals was added to a trial-by-trial ideal observer limited to 36 dots and integration intervals ≤ 165 ms. The model fit human data remarkably well across all three experiments with no free parameters. Conclusions: SBF is based on recovery of local orientation from mechanisms that sample, subject to encoding imprecision of relevant display properties, pairs of element changes under specific spatial and temporal constraints.
Meeting abstract presented at VSS 2015
This PDF is available to Subscribers Only