The intent of this project is to play around with pen-and-ink illustrations (like figure 1). Instead of generating them interactively, as described in the papers, I would like to generate them automatically from images.
My last presentation showed the results of the Marr Hildreth edge detector,
which produced results that weren't exactly what I had in mind. So, I now
detect edges based on the CIELab color difference metric.
Additionally, the edge break down into two (non-exclusive) categories:
color boundaries and brightness boundaries. If we only look at color
boundaries we get:
Based on these outlines, the big question is how to shade the interior of the
shapes defined by the outlines. I've written code to color rectangular
regions of the image while respecting boundaries:
The outlines need to be thinned and turned into real curves (rather then just collections of points). This should allow for simpler segmentation and help determine the direction of lines for pen-and-ink style shading.
More stylization of the edges based on the curve information.
A real user interface.