Purchase this article with an account.
Stephan Tschechne, Heiko Neumann; The Structure of Optical Flow for Figure-Ground Segregation. Journal of Vision 2012;12(9):242. doi: https://doi.org/10.1167/12.9.242.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Introduction: Patterns of accretion and deletion of visual structure are indicative of ordinal depth structure along motion boundaries [Gibson et al., Perception & Psychophysics 5, 1969]. We propose a neurally inspired model that derives ordinal depth order by analyzing spatio-temporal configurations of optic flow patterns at and the motion of kinetic boundaries.
Method: A neural model architecture is proposed that detects and spatially integrates motion signals in accordance with cortical stages V1 and MT. Changes in flow pattern characteristics (flow discontinuities or speed and direction changes) are detected in model MST using a mechanism to signal boundary orientations. Furthermore, we suggest that neural mechanisms exist that are sensitive to register secondary motion features, ie. that of moving kinetic boundaries. Activity of such kinetic boundary movement in combination with signals from model MST are used to generate border ownership signals which can be interpreted as likelihoods of a contour that belongs to a surface in a certain direction. An integration process operating at lower spatial resolution provides feedback to the border ownership signal and allows Gestalt-like grouping of boundary pairs with opposite ownership direction tags.
Results: We probed our model with various artificial scenes and with a benchmark consisting of real-world sequences [Stein & Hebert, IJCV 82, 2009]. The model generates stable border ownership signals from interactions of the hierarchical stages of the model. We particularly demonstrate that inconsistencies in the movement of kinetic boundaries compared to the surrounding (discontinuous) flow pattern lead to perceptual ambiguities.
Conclusion: Our proposed model shows that local depth structure can be computed by local neural mechanisms in a distributed fashion. We show that local estimation of motion gradient and movement of kinetic boundaries is sufficient to estimate ordinal depth structure.
Meeting abstract presented at VSS 2012
This PDF is available to Subscribers Only