September 2011
Volume 11, Issue 11
Free
Vision Sciences Society Annual Meeting Abstract  |   September 2011
Ordinal depth from occlusion using optical flow: A neural model
Author Affiliations
  • Stephan Tschechne
    Ulm University, Institute for Neural Information Processing, 89069 Ulm, Germany
  • Heiko Neumann
    Ulm University, Institute for Neural Information Processing, 89069 Ulm, Germany
Journal of Vision September 2011, Vol.11, 716. doi:10.1167/11.11.716
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Stephan Tschechne, Heiko Neumann; Ordinal depth from occlusion using optical flow: A neural model. Journal of Vision 2011;11(11):716. doi: 10.1167/11.11.716.

      Download citation file:


      © ARVO (1962-2015); The Authors (2016-present)

      ×
  • Supplements
Abstract

Problem. An occluding surface boundary is created by the spatio_temporal pattern of deletion and accretion of visual structure in which ordinal depth relations along the edge are defined by the arrangement of occluding and occluded surface regions (Gibson et al., Perception & Psychophysics 5, 1969). What are the neural computational mechanisms underlying the motion_based segregation of figure from ground based on occlusion boundary information?

Method. We propose a biologically inspired model of cortical motion detection and integration. Local motion signals are detected in model V1 which are spatially integrated over a larger neighborhood in model MT. Top_down feedback signals from MT to V1 disambiguate and stabilize the local measurements. The presence of temporal occlusions generated by surfaces hovering at different depths is evidenced by a transition from high motion energy when coherent motion is measured to low motion energy in cases of failure to find coherent spatio_temporal structure. In cases of temporal dis_occlusions low motion energy responses change into high motion energy response patterns. We propose an opponent scheme of temporal on/off interactions in which local motion energy signals from model V1 are spatially integrated by the temporally offset on/off subfields. This mechanism is gated by direction selective model MT cells to make the spatio_temporal occlusion detection selective to different motion directions.

Results. The model was probed with artificial scenes of moving and mutually occluding object surfaces. Motion is estimated by the model and activities at the stage of occlusion detection correctly indicate the foreground objects when they partially occlude the background. The occlusion/dis_occlusion boundary responses together with directional motion signals determine the border-ownership direction of an occluding surface. This demonstrates that spatio-temporal figure-ground separation can be achieved by local mechanisms at early and intermediate stages of the dorsal visual pathway.

Supported in part by 7th framework program ICT-project no. 215866-SEARISE and Transregional Collaborative Research Center SFB/TRR 62 Companion-Technology for Cognitive Technical Systems funded by the German Research Foundation (DFG). 
×
×

This PDF is available to Subscribers Only

Sign in or purchase a subscription to access this content. ×

You must be signed into an individual account to use this feature.

×