How to detect dominant points on 3-D curves

A method for detecting dominant points on a 3-D digital curve is presented. The essential point is how accurately local properties such as curvatures and tangent vectors at every point are computed. This procedure first computes temporary properties by line or circle approximation using k-neighbors, then determines the optimum approximation by comparing the error value and correcting properties, and finally only for the point which has a large error value, adapts a threshold value K which defines the neighboring points for approximation. This method can detect not only corners but inflection points where two curves which have different convexities smoothly join and transition points where a straight line and a curve connect smoothly.