Border-ownership computation

neural mechanisms underlying emeregent perceptual organization


In collaboration with Dr. Vicky Froyen, we developed neuro-computational model to reproduce human perception of "border-ownership" (BOWN). It modelled the BOWN sensitive neurons first reported by Zhou et al. (2000) J. Neurosci. The algorithm reflect global coherence of the ownership signals and contrast polarities. As the results, the model shows extremely robust responses corresponding human perception of figure-ground organization.

This work is now published in Psychological Review (here).

Background: The visual system performs remarkably well to perceive the depth order of regions without stereo disparity. This indicates the importance of figure-ground organization based on pictorial cues. To understand how figure-ground organization emerges through image signal processing, it is essential how the global configuration of the image is reflected. In the past, many neuro-computational models to reproduce figure-ground organization implemented algorithms to give a bias to convex shapes. However, in certain conditions, non-convex regions can be perceived as figural, e.g. a convex region can be perceived as a hole. The essential difference is that the surface properties of the hole region are consistent with the background and hence they are grouped together in our perception. We argue that the long-range consistency of surface properties is reflected in the computational processes. It is possible that a class of border-ownership-sensitive neurons that are also sensitive to contrast polarity underlies this process. Furthermore, we argue that T-junction detection is not necessary to reflect occlusion conditions if surface consistency is detected instead. 

Our latest work of BOWN computation by the new DISC model (DISC 2) is published in Psychological Review (2021, here).