Stylization and Abstraction of Photographs


 These images have been rendered from photographs in a NPR style. Perhaps more interestingly their content has been simplified, removing detail a particular viewer found unimportant. The interaction for this is done using an eye tracker, the user simply looks at the image, identifying important regions naturally and effortlessly.   

The official page for this which has PDFs of the article in Siggraph and a related article in NPAR as well as more results. http://www.cs.rutgers.edu/~decarlo/stylize.html

On this page I've put some more results to give a better idea of what the system can do, you can look at the page above for side by side comparisons between input and output.


Something a little different here,  thresholding the region colors after generating the image creates a surprisingly convincing comic book effect on high contrast images.

Video is a natural extension but a bit tricky from eye tracking interaction viewpoint, and has usual temporal coherence problem. For the fun of it, a little clip  using the thresholded comic style that addresses neither of these issues, no coherence is attempted and eye tracking isn't used to control detail, but shows some of the possibilities.