Curvilinear feature detection from curvature estimation

Abstract

The DOS/sup +/ method, a particular case of the difference of slopes (DOS) approach, is shown to be effective for estimating curvature in terms of signal detectability. Relationships are established between the DOS/sup +/ parameters and feature parameters to distinguish corner and curve features and to accurately determine the locations of corners, transitions between straight lines and curves, and curve centers. Examples show the results of this method applied to curvilinear features and diagrams.<>