Highly accurate segmentation using geometric attraction-driven flow in edge-regions
Jooyoung Hahn, Chang-Ock Lee · University of Minnesota Digital Conservancy (University of Minnesota) · 2006
A highly accurate segmentation algorithm is proposed for extracting objects from an image that has simple background colors or simple object colors. Two main concepts, geometric attraction-driven flow (GADF) and edge-regions are combined to detect exact boundaries of objects in a sub-pixel resolution. GADF gives exact locations of boundaries and edge-regions help to make initial curves close to objects. For high accuracy in segmentation, we additionally propose a local region competition algorithm which detects perceptible boundaries of objects. The whole algorithm is able to extract objects even though there are weak edges, shadows, and highly non-convex shapes. Furthermore, there are no manipulations of parameters in the whole process.