Pictograms (icons) are ubiquitous in visual communication, but creating the best icon is not easy: users may wish to see a variety of possibilities before settling on a final form, and they might lack the ability to draw attractive and effective pictograms by themselves. We describe a system that synthesizes novel pictograms by remixing portions of icons retrieved from a large online repository. Depending on the user's needs, the synthesis can be controlled by a number of interfaces ranging from sketch-based modeling and editing to fully-automatic hybrid generation and scribble-guided montage. Our system combines icon-specific algorithms for salient-region detection, shape matching, and multi-label graph-cut stitching to produce results in styles ranging from line drawings to solid shapes with interior structure.

@inproceedings{Liu:2016:DI,
   author = "Yiming Liu and Aseem Agarwala and Jingwan Lu and Szymon Rusinkiewicz",
   title = "Data-Driven Iconification",
   booktitle = "International Symposium on Non-Photorealistic Animation and Rendering
      (NPAR)",
   year = "2016",
   month = may
}