Global Curvature Estimation for Corner Detection

Abstract

The paper starts with presenting three curvature estimators which follow definitions (approaches) in differential geometry. Digital-straight segment (DSS) approximation is used in those estimators, we point to problems caused by this approach, and propose simple ways for eliminating those problems. The paper then informs about multigrid analysis experiments, where all estimators appear to be multigrid convergent when digitizing an ellipse. The paper also applies these estimators for corner detection and compares their performance with a recently published heuristic corner-detection approach by means of multigrid analysis. Experiments indicate that corner detectors (based on curvature estimation) perform about as good as the heuristic method for large grid resolutions, and one detector might be even superior.