Fidelity vs. Simplicity: a Global Approach to Line Drawing Vectorization

摘要

Vector drawing is a popular representation in graphic design becausernof the precision, compactness and editability offered by parametricrncurves. However, prior work on line drawing vectorizationrnfocused solely on faithfully capturing input bitmaps, and largelyrnoverlooked the problem of producing a compact and editable curvernnetwork. As a result, existing algorithms tend to produce overlycomplexrndrawings composed of many short curves and controlrnpoints, especially in the presence of thick or sketchy lines that yieldrnspurious curves at junctions. We propose the first vectorization algorithmrnthat explicitly balances fidelity to the input bitmap withrnsimplicity of the output, as measured by the number of curves andrntheir degree. By casting this trade-off as a global optimization,rnour algorithm generates few yet accurate curves, and also disambiguatesrncurve topology at junctions by favoring the simplest interpretationsrnoverall. We demonstrate the robustness of our algorithmrnon a variety of drawings, sketchy cartoons and rough designrnsketches.