Aesthetics of curvature bases for sketches

Abstract

In this work we propose a curve approximation method that operates in the curvature domain. The curvature is represented using one of several different types of basis functions (linear, quadratic, spline, sinusoidal, orthogonal polynomial), and the curve's geometry is reconstructed from that curvature basis. Our hypothesis is that different curvature bases will result in different aesthetics for the reconstructed curve. We conducted a user study comparing multiple curvature bases, both for aesthetics and similarity to the original curve, and found statistically significant differences in how people ranked the reconstructed curve's aesthetics and similarity. To support adaptive curve fitting we developed a fitting algorithm that matches the original curve's geometry and explicitly accounts for corners.