Interest
New Inking Technology from Waseda University Makes Line Art Easy
posted on by Eric Stimson
Inking rough sketches is one of the necessary steps in converting a simple drawing into a polished showpiece work. Although there is commercial software to ink line art automatically, a research team from Tokyo's Waseda University led by Edgar Simo-Serra and Satoshi Iizuka has developed new technology that arguably produces a cleaner, more polished final product.
The technology uses a fully convolutional neural network. Images are shrunk ("down-convoluted"), processed ("flat-convoluted") and then enlarged to the original resolution ("up-convoluted"). It uses raster images like pencil sketches as input, and is able to process images of any aspect ratio and dimension.
The team claims that 97% of users prefer their method over pre-existing ones. The image below compares their results with those of Potrace and Adobe Illustrator's Live Trace function; the former is considered to have too many extraneous lines, while the latter is judged to over-simplify the drawing.
If you're interested, you can read Simo-Serra and Iizuka et al's paper, "Learning to Simplify: Fully Convolutional Networks for Rough Sketch Cleanup" here. The team is also exploring a technique for colorizing century-old black-and-white photos using a similar fully convolutional neural network.
Sources: Gigazine, Waseda University and The Verge