From visualization to perceptual organization

A key initial stage in the process of translating sensor data into symbolic representations is perceptual organization where edge pixels or voxels are grouped into a set of disjoint segments. For example, edge pixels are linked to form contour lines that can be associated with boundaries of regions; or edge voxels are grouped to form surfaces that often represent boundary surfaces of volumes. This paper studies the perceptual organization problem in both two and three dimensions. An interdisciplinary research work is presented which explores the synergy between perceptual organization in computer vision and streamline and iso-surface techniques in visualization. We show how such visualization techniques can be adapted and unified into a disjoint-union-find framework to solve the perceptual organization problems. The proposed algorithms have been successfully applied to many real-world 2D and 3D images.<>