Segmentation of Edges in 3 - D Object Space

In our feature-based matching approach, zero-crossings are matched and represented in 3-D object space by a sequence of densely spaced points. These spatial curves form the basis for reconstructing the surface. Since edges are likely to correspond to object boundaries, the 3-D curves also serve as an important input for object recognition. In this paper we address the problem of segmenting the contours in straight lines and regular curves. We compare different methods, such as split-andmerge and a 3-D version of the 'I/J S method.