Abstract
This paper describes a new method to extract features of contour of a figure using its curvature changes. The local curvature maxima and minima along with the contour line are powerful features to characterize the contour of a figure, which correspond to convex and concave vertices, respectively. While the curvature is originally defined in terms of derivatives, for a digital figure, they are defined by using differences of line directions at two points with an appropriate distance scales, because a digital curve is given as a series of points.
However, when some vertices locally neighbour on the contour line, false curvatures are calculated with some range of the scales. Where two convex vertices nearly neighbour, for example, a false larger maximal value is calculated if too large scale is used. But, employing some multiple scales and comparing the results, such false curvature maxima appear to change their values in peculiar forms according to the locations and angles of vertices included in that part.
In this method, a set of curvatures with various scales, which is called a multi-scale curvatures, is calculated. Analyzing the curvature changes with respect to the multiple scales, therefore, the features of the original contour, such as the locations and angles of vertices, are extracted from the multi-scale curvatures. The “view-field resolution” for the analysis of features of figures are determined automatically, for one by one figure and its part.
