Understanding and simulating border ownership centered segmentation

Border ownership is a critical component of biological vision systems, enabling object segmentation and figure-ground organization. Understanding border ownership-centered segmentation is essential for advancing vision neuroscience and computer vision. In our prior work, "Border Ownership, Category Selectivity, and Beyond" (Chen et al., 2022)⁶, we introduced a channel-based representation for coding border ownership and category selectivity. Building on this foundation, this paper addresses the remaining challenges in border ownership-centered segmentation. We explore binocular disparity representation, the generation of binocular border ownership for contrast- and disparity-defined objects, and border ownership generation for illusory and contour-defined objects, including the Rubin Face-Vase illusion. We also propose the creation of 'object pointers' through hypothetical active neurons to address the 'surface filling-in' process in neuroscience and generate 'instance masks' in computer vision. Finally, synthesizing these findings, we present a comprehensive model for border ownership-centered figure-ground organization. This model integrates global context awareness, distributed processing, and multi-cue representations, bridging gaps between biological vision and computational applications.