Multi-scale contour segmentation

Effective local segmentation of contours is an important problem which arises in occluded object recognition as well as other areas. For any recognition system to perform successfully, the segmentation procedure used must be robust in presence of noise and local distortions of shape. Furthermore, it should be based on geometric invariants so that the segmentation will not be affected by arbitrary choices. This paper proposes a new multi-scale segmentation routine for planar contours which is based on the curvature scale space representation. Curvature zero-crossing segments extracted from a continuum of scales are utilized for robust segmentation of planar curves. Curvature zero-crossing points are effectively tracked across increasing scales to ensure that the same segment is extracted only once. This approach is more robust than techniques which try to recover features/segments from a stable scale and, as a result, risk over- or under-segmentation of the input contour.