PolyFit: Perception-Aligned Vectorization of Raster Clip-Art via Intermediate Polygonal Fitting

摘要

Raster clip-art images, which consist of distinctly colored regions separatedby sharp boundaries typically allow for a clear mental vector interpretation.Converting these images into vector format can facilitate compactlossless storage and enable numerous processing operations. Despite recentprogress, existing vectorization methods that target such data frequentlyproduce vectorizations that fail to meet viewer expectations. We presentPolyFit, a new clip-art vectorization method that produces vectorizationswell aligned with human preferences. Since segmentation of such inputsinto regions had been addressed successfully, we specifically focus on fittingpiecewise smooth vector curves to the raster input region boundaries, a taskprior methods are particularly prone to fail on. While perceptual studiessuggest the criteria humans are likely to use during mental boundary vectorization,they provide no guidance as to the exact interaction between them;learning these interactions directly is problematic due to the large size of the solution space. To obtain the desired solution, we first approximate theraster region boundaries with coarse intermediate polygons leveraging acombination of perceptual cues with observations from studies of humanpreferences. We then use these intermediate polygons as auxiliary inputsfor computing piecewise smooth vectorizations of raster inputs. We define afinite set of potential polygon to curve primitive maps, and learn the mappingfrom the polygons to their best fitting primitive configurations fromhuman annotations, arriving at a compact set of local raster and polygonproperties whose combinations reliably predict human-expected primitivechoices. We use these primitives to obtain a final globally consistent splinevectorization. Extensive comparative user studies show that our methodoutperforms state-of-the-art approaches on a wide range of data, where ourresults are preferred three times as often as those of the closest competitoracross multiple types of inputs with various resolutions.