Segmentation of digitized planar curves is one of the most important elements in early image processing, because a segmented image can describe the object profile in a compact form to facilitate higher level vision processing. In many applications, it is necessary to decompose an object boundary contour into several primitives, such as segments and curves. In this paper, a two-stage hybrid technique for the segmentation of two-dimensional (2D) curves is presented, in which the number of segments is assumed to be known. First, the boundary is iteratively approximated using a split-and-merge method. Next, an end-point adjustment procedure is applied to reach the best-fitting polygonal approximation. A computational comparison with two existing methods shows that the proposed technique is fast and accurate. An application of the new segmentation technique to industrial part inspection is also provided.
All Science Journal Classification (ASJC) codes
- Signal Processing
- Computer Vision and Pattern Recognition
- Artificial Intelligence