Creating software that can understand the range of sketches that people produce is a challenging problem. One source of difficulty is that people often include textures in their drawings. This paper shows how to use Ising models, a technique from computer vision at the level of pixels, for decomposing digital ink into a hierarchy of edge-based structures that provide more concise qualitative representations of textures in hand-drawn sketches. We analyze the compression efficacy, strengths and weaknesses of this qualitative representation technique using a subset of a large-scale sketch corpus.
|Original language||English (US)|
|Number of pages||9|
|State||Published - 2015|