Significant point detection and boundary representation with lines and circular arcs, using a scanning box method

A boundary curve is represented by circular arcs and lines to obtain explicit geometrical features. Corner points, which are local extremum points, are detected by the symmetry property of the points. An inflection point is a zero-crossing point between two neighboring extremum points. In the LMS (least-mean-square) sense, a circular arc between two neighboring significant points is optimized. The similarity of neighboring segments is checked in radius and center, and the segments are merged when the difference is negligible. The representation result obtained here is a good approximation of the original curve, in the sense that it makes critical points explicit and achieves error-minimized representation.<<ETX>>