Selective extraction of curvilinear edges using gradient angle histograms

Detecting curvilinear edges, that includes curved or linear or both, is a crucial image processing operation in applications like lane detection, shape/object detection, industrial shape detection, iris detection etc. In this paper, we propose a novel technique to generate edge maps containing specific kinds of edges. Given an edge map, the proposed technique uses gradient angle histograms (GAHs) to: (1) selectively extract either linear or curved edges from a given edge map and, (2) extract curved edges of varying curvatures. The GAH signature of various features are studied, and the relationship between the curvature of the curves of interest and the different parameters in a GAH, are described. Experiments on real images show that given a set of initialization conditions, the proposed technique can be used effectively extract edges of different curvatures.