Construction of a polygonal model using the Fisher-ratio criterion

A method is described to construct a polygonal model for a family of objects, consisting of various classes. The model optimally discriminates between classes on the shape of the contour. The method assumes some characteristic points (landmarks) on the contour are available for comparing corresponding curves over all objects. Relating a feature (angle of directional change) to each point on the curve makes it possible to select new characteristic points using the ratio of variation between and within classes. Two approaches are proposed for successively selecting the characteristic points, a recursive and a multivariate one. The method is the first automatic polygonization method which takes into account information of multiple objects. It is not optimized for a single object, but for discrimination between classes of similar objects with local shape differences.