Multilayer cooperative/competitive method for creating hierarchical structures by clustering maximally-related nodes


A cooperative/competitive method can be used to generate hierarchical structures for clustering problems. These structures can prove useful in achieving high-level machine vision goals such as scene analysis through perceptual organization. The cooperative/competitive method was specifically designed for clustering related image segments to create hierarchical scene structures. The method addresses the need for rapid extraction of clusters from large sets of regions with different inter-region relationships. The multiple relationships cooperate to form clusters from the input nodes. These clusters then compete, eliminating all but the strongest. The input nodes that comprise the surviving clusters are combined. This process is applied repeatedly to produce a full hierarchical structure. A specific example uses this approach to create a hierarchical structure from a synthetic segmented image.

This document is currently not available here.