A multilayer cooperative/competitive method for grouping and organizing related image segments

Most high-level machine vision systems, perform poorly when confronted with real world images. Occlusions, specularities and actual physical differences between segments cause objects or related information to be partitioned into multiple segments. This thesis presents a methodology for joining together related regions and creating an organized representation of the image. This methodology uses perceptual features and a cooperative/competitive process to group regions and produce perceptually-based hierarchical structures. The hierarchical structures overcome many difficulties associated with segmentation. In this thesis hierarchical structures are created for three images. In all three examples, regions separated by occluding objects or faulty segmentation were grouped together. The organized representation of an image created by this methodology will be a valuable tool in future vision systems and is one step towards more powerful vision capabilities .