Generic Edge Tokens: Representation, Segmentation and Grouping

This paper presents a perceptual organization based method for describing, extracting and grouping generic edge features, called Generic Edge Tokens (GET). A GET is a perceptually significant image primitive which represents a class of qualitatively equivalent structure elements. A complete set of GETs includes both linear and non-linear segment classes and junctions of the segments. Edge traces extracted from image are segmented into GETs according to Gestalt Laws of proximity, continuity, and similarity in terms of descriptive image geometry and edge strength. In grouping 2D shapes, an object is described by its made-up GETs and the relations of GETs. The examples of ellipses and parallelograms groupings are provided as an illustration of the method.