Abstract
Curvature detection is important particularly in shape representation and object detection. Curvature contains information related to the features of an object, and a specific object can be extracted from an image so long as the features of the borderline can be determined from its curvature. Besides, accurate curvature detection can make applications such as checking of objects possible and industrial applicable. This paper proposes a novel mechanism for edge segment based curvature detection : a Laplacian and Gaussian (∇2G) filter is applied to an input image, and edge segments of the image are detected first ; By searching for peaks in positive and negative ∇2G values obtained from the edge segments, the borderline of an object in edge-segment units can be easily obtained, and the curvature can be simultaneously detected from the peak distance. With this method, high accurate curvature detection can be realized, and the feature shape of edge segments can be classified. A real life case study also demonstrates the feasibility and usefulness of the proposed method.
