Beitraege zur automatischen Konturextraktion und Formerkennung.

A simple criterion is proposed for the selection of contour points as a suitable representation of intensity changes in images from a hierarchy of filter responses. The operators are the first derivatives of 2-dimensional Gaussians with different widths whose positive and negative parts are separately normalized to +1 and -1, respectively. The resulting gradient vectors of the image function give the direction of maximal intensity changes with high resolution. Using this information, closed chains of contour points are obtained which yield a good image segmentation. Corner points of regions are used for an invariant shape description. (IITB)