Layered Disparity Representation and Active Neurons for Enhanced Depth Perception, Border Ownership Generation, and Surface Filling-In

A method for representing depth in image or vision processing, comprising:

• (a) receiving binocular image data;
• (b) representing near-distance disparities using a mapping consistent with Epipolar geometry;
• (c) representing far-distance disparities using a logarithmic disparity mapping;
• and
• (d) combining said near-distance and far-distance disparity representations to form a layered disparity representation across multiple depth ranges.

A system for figure-ground segmentation in image or vision processing, comprising:

• (a) a plurality of computational units (“active neurons”) configured to receive disparity or image input;
• (b) said active neurons being configured to propagate border ownership signals along connected contours;
• (c) said active neurons being further configured to spread ownership signals into surface regions for surface filling-in;
• and
• (d) wherein said active neurons maintain object identity during propagation across contours and surfaces.

A system for visual scene segmentation, comprising:

• (a) a disparity module configured to represent depth using a layered disparity representation, said layered disparity representation including Epipolar disparity for near-distance depths and logarithmic disparity for far-distance depths, wherein said logarithmic disparity flattens the magnitude difference between small and large disparities;
• (b) a border ownership module configured to generate border ownership;
• (c) an active neuron module configured to propagate ownership signals along contours and to perform surface filling-in;
• and
• (d) wherein said disparity module, border ownership module, and active neuron module are integrated to provide a unified figure-ground segmentation of a visual scene.