A Unified Mechanism for Depth Perception and Figure-Ground Organization

Abstract

Understanding how the visual system unifies depth perception and figure-ground organization remains a central challenge in vision science. Stereoscopic disparity and border ownership have long been treated as distinct processes, yet both arise with similar timing in area V2 and exhibit strikingly parallel properties. Here we propose a unified mechanism that integrates both within a single hierarchical framework. Building upon the layered disparity representation introduced in our prior work, we show that border ownership--serving as the key component of figure- ground organization--emerges intrinsically from thresholded relative-disparity differencing, providing a unified basis for contour-based segmentation and depth perception. Using biologically constrained network models (TcNet), we demonstrate that this mechanism generalizes across binocular and illusory depth domains, bridging physical and non-epipolar disparity cues. At the systems level, we identify complementary dual feedback pathways from area V4: V4[->]V1, which refines and re-weights layered disparity maps, and V4[->]V2, which integrates global contextual priors such as shape continuity and category consistency. Together, these recurrent loops balance fine-scale depth precision with large-scale figure-ground coherence, offering a unified explanation for perceptual alternations such as the Rubin Face-Vase and Kanizsa illusions. This framework integrates intrinsic disparity computation, border-ownership dynamics, and hierarchical feedback modulation into a coherent, biologically plausible model. It provides a mechanistic account of how early visual areas interact with higher cortical feedback to achieve stable yet flexible perception of depth and figure-ground organization within a unified temporal cycle.