A curvature estimation for pen input segmentation in sketch-based modeling

Authors: Dae Hyun Kim , Myoung-Jun Kim

Comments:

1. Daniel's blog

Summary:

This paper proposes an algorithm to find the corners based on the curvature and direction measures. Preprocessing steps include resampling strokes to make the inter point distance constant. This implies the curvature can be estimated only based on the direction values of consecutive points by removing the constant distance term.

Local Convexity: Curvature at a point is defined as the sum of curvature values over a window of k points before and after the current point (support window). This sum is greater if the point is a corner. This fails when there are corners which are close together.

Local Monotocity: When the Local convexity calculation fails, monotonocity is used to find the corners. This is used where corners are close together. The algorithm finds if the points around a corner is monotonously increasing or decreasing in curvature and finds the minima in the support window .This minima is then stated as the corner.

Discussion:

This method works well for polylines and smoother curves. Like any other corner finding algorithm, It fails for small curves and curves with irregularly changind curvature.