Edge detection using two-dimensional local structure information

Local intensity discontinuities, commonly referred to as edges, are important attributes of an image. Many imaging scenarios produce image regions exhibiting complex two-dimensional (2D) local structure, such as when several edges meet to form corners and vertices. Traditional derivative-based edge operators, which typically assume that an edge can be modeled as a one-dimensional (1D) piecewise smooth step function, give misleading results in such situations. Leclerc and Zucker introduced the concept of local structure as an aid for locating intensity discontinuities. They proposed a detailed procedure for detecting discontinuities in a 1D function. They had only given a preliminary version of their scheme, however, for 2D images. Three related edge-detection methods are proposed that draw upon 2D local structural information. The first method greatly expands upon Leclerc and Zucker's 2D method. The other two methods employ a mechanism similar to that used by the maximum-homogeneity filter (a filter used for image enhancement). All three methods permit the detection of multiple edges at a point and have the flexibility to detect edges at differing spatial and angular acuity. Results show that the methods typically perform better than other operators.