StrokeAggregator

Abstract

When creating line drawings, artists frequently depict intended curves using multiple, tightly clustered, or overdrawn, strokes. Given such sketches, human observers can readily envision these intended, aggregate , curves, and mentally assemble the artist's envisioned 2D imagery. Algorithmic stroke consolidation---replacement of overdrawn stroke clusters by corresponding aggregate curves---can benefit a range of sketch processing and sketch-based modeling applications which are designed to operate on consolidated, intended curves. We propose StrokeAggregator , a novel stroke consolidation method that significantly improves on the state of the art, and produces aggregate curve drawings validated to be consistent with viewer expectations. Our framework clusters strokes into groups that jointly define intended aggregate curves by leveraging principles derived from human perception research and observation of artistic practices. We employ these principles within a coarse-to-fine clustering method that starts with an initial clustering based on pairwise stroke compatibility analysis, and then refines it by analyzing interactions both within and in-between clusters of strokes. We facilitate this analysis by computing a common 1D parameterization for groups of strokes via common aggregate curve fitting. We demonstrate our method on a large range of line drawings, and validate its ability to generate consolidated drawings that are consistent with viewer perception via qualitative user evaluation, and comparisons to manually consolidated drawings and algorithmic alternatives.

A benchmark for rough sketch cleanup

“…They do this by clustering strokes and then replacing each cluster with a single representative curve. Numerous later works have been proposed following this same basic cluster-and-replace framework for vector graphics [Liu et al 2018[Liu et al , 2015Ogawa et al 2016;Orbay and Kara 2011;Shesh and Chen 2008]. Another line of work simultaneously vectorizes and simplifies a raster image of a sketch [Bessmeltsev and Solomon 2019;Donati et al 2019;Favreau et al 2016;Kim et al 2018;Noris et al 2013;Parakkat et al 2018].…”

“…Four methods take raster input and produce vector output by applying heuristic-based optimization: PolyVector [Noris et al 2013], FidelitySimplicity [Favreau et al 2016], Topolo-gyDriven [Bessmeltsev and Solomon 2019], and DelaunayTriangulation [Parakkat et al 2018]. One method, StrokeAggregator [Liu et al 2018], takes vector input and produces vector output by applying perceptual principles . Two data-driven approaches based on convolutional neural networks, MasteringSketching [Simo-Serra et al 2018a] and its follow-up RealTimeInking [Simo-Serra et al 2018b], take raster input and are the only two methods which produce raster output.…”

“…We use the authors' own implementations of their algorithms. We also evaluated two pipelines: one of two selfdescribed vectorization algorithms, TopologyDriven [Noris et al 2013] or PolyVector [Bessmeltsev and Solomon 2019], followed by StrokeAggregator [Liu et al 2018], the cleanup method that requires vector input.…”

“…To account for this, we evaluated raster images at their original resolution, thresholded images at the same resolution and resized to have 1000 and 500 pixels along their long edge, and vectorized images rasterized at 1000-and 500-pixel dimensions. We evaluate StrokeAggregator [Liu et al 2018] only on the professionally vectorized inputs (all layers and shape strokes only).…”

“…Such algorithms have been explored in the past [Barla et al 2005;Orbay and Kara 2011]. In the last five years, there has been a flurry of work [Bessmeltsev and Solomon 2019;Favreau et al 2016;Liu et al 2018Liu et al , 2015Noris et al 2013;Parakkat et al 2018;Simo-Serra et al 2018a,b]. These works differ substantially in the assumptions they make on the input and output sketches.…”