One-Dimensional Edge Detection and Representation

Abstract

One reason no single edge detector has been found that satisfies the requirements of a wide range of image analysis applications is that existing edge detectors often produce unjustifiable interpretations of edge regions without sufficient contextual or external information. One- dimensional edge semantics are examined in detail, and a one-dimensional edge-detector is proposed whose output is not just a local edge point assertion but a data structure. This data structure contains alternative interpretations of edge crossections as global gray level transition regions of variable widths. The detector is parameter-free, and a set of descriptive edge features is proposed.