Scene di usion strength di usion controlFigure 1: Diffusion Curves allow us to draw vectorial images with a rich set of color gradients (left). It is based on a diffusion process that propagates color information from curves in the scene. While the colors could be chosen arbitrarily, the diffusion itself is not controllable by the user. Our work introduces ways to alter diffusion behavior. This allows us to reduce the number of color definitions for an equivalent output (middle), to control the diffusion strength of certain colors (right, floor), or even influence diffusion directions (right, cushion).

Region-based approaches to cel painting typically use shape similarity and topology relations between regions of consecutive animation frames. This paper presents a new colorization algorithm based on topological differences defined over a hierarchical graph of adjacent regions, which allows an almost full automatic colorization process. Also this paper discusses other attributes that improve the solution of the image association problem.

In art, grouping plays a major role to convey relationships of objects and the organization of scenes. It is separated from style, which only determines how groups are rendered to achieve a visual abstraction of the depicted scene. We present an approach to interactively derive grouping information in a dynamic 3D scene. Our solution is simple and general. The resulting grouping information can be used as an input to any "rendering style".We provide an efficient solution based on an extended mean-shift algorithm customized by user-defined criteria. The resulting system is temporally coherent and real-time. The computational cost is largely determined by the scene's structure rather than by its geometric complexity. This is the authors' version of this paper. The definitive one was published in the NPAR 2008 Proceedings.

Shading for cel animation based on images is a recent