Non-photorealistic rendering from photographs most often uses “filters” that change a small image patch into a brush stroke. But making brush strokes can never transform a photograph into art that requires objects to change shape: cubism is an obvious example.
Our cubism project was the first to move away from using filters and towards image understanding. Three advances were necessary: (1) We defined pixel salience as global property - that is a pixel is important only when compared to all other pixels in the image; see Salient Painting. (2) We aggregated salient pixels into image regions to be used as painting primitives. (3) We composited a painting by cutting salient regions from many photographs.
This work has been featured in press across the world: the Times, the Times Higher Education Supplement, The Toronto Globe and Herald, the Seattle Intelligensia, and many news web-sites. Academic publications include.
Papers
J.Collomosse and P. Hall, “Cubist Style Rendering from Photographs”, IEEE Transactions on Visualization and Computer Graphics, 9(4), 443-453, 2003.
