Comparison and improvement of tangent estimators on digital curves

Abstract

Many contour-based applications rely on the estimation of the geometry of the shape, such as pattern recognition or classification methods. This paper proposes a comprehensive evaluation on the problem of tangent estimators on digital curves. The methods taken into account use different paradigms: approximation and digital geometry. In the former paradigm, methods based on polynomial fitting, smoothing and filtering are reviewed. In the latter case of digital geometry, we consider two methods that mainly rely on digital straight line recognition [J.-O. Lachaud, A. Vialard, F. de Vieilleville, Fast, accurate and convergent tangent estimation on digital contours, Image and Vision Computing 25(10) (2007) 1572–1587] and optimization [B. Kerautret, J.-O. Lachaud, Robust estimation of curvature along digital contours with global optimization, in: Proceedings of Discrete Geometry for Computer Imagery, Lyon, France, Lecture Notes in Computer Science, vol. 4992, Springer, Berlin, 2008]. The comparison takes into account objective criteria such as multi-grid convergence, average error, maximum error, isotropy and length estimation. Experiments underline that adaptive methods based on digital straight line recognition often propose a good trade-off between time and precision and that if precision is to be sought, non-adaptive methods can be easily transformed into adaptive methods to get more accurate estimations.