More Optimal Strokes for NPR Sketching

John Lewis
Nickson Fong
Xie XueXiang
Stanford University
Nanyang Technological University
Seah Hock Soon
Tian Feng
Nanyang Technological University
Nanyang Technological University


"Sketching" is a drawing process that displays successive approximation: the scene is first coarsely rendered, then successively more precise strokes are overlaid to improve the rendition. Edges and contours are segmented into a number of (often overlapping) strokes, with shorter strokes characteristically chosen in high-curvature regions. In this paper we approximate the feel and strategy of such contour stroking.

Simple curvature-based segmentation schemes (such as breaking contours at curvature maxima) will not always work for this purpose, both because the initial contour description may contain high curvature artifacts from rasterization and noise, but more importantly because sketching initially ignores small details regardless of whether they correspond to high-curvature locations on a contour.

We approach this problem with a perceptual segmentation scheme that combines an affinity measure incorporating perceptual continuation with recent segmentation algorithms. The resulting algorithm produces strokes that strongly resemble those chosen by trained artists. In addtion, it has a scale parameter than can be scheduled to produce the coarse-to-fine successive approximation seen in sketching.

Graphite 2005 paper:
More Optimal Strokes for NPR Sketching

(Left) breaking silhouettes at curvature maxima works poorly if there are small details, noise, or rasterization artifacts on the contour. (Right) Strokes obtained using perceptually oriented segmentation break at more salient points.