The Biological Simulator Behind “Cellular Texture Generation”

The simulator used to produce the images in the Siggraph 95 paper “Cellular Texture Generation” [5] was not originally designed as a Computer Graphics particlesystem simulator, although there are many similarities. It is instead a simulator designed for studying pattern formation in early biological development. Because it is intended for studying the patterns arising from the interactions of cells, we naturally focus on the issues like contact between cells, cell proliferation, etc. When we decided to apply this simulator to creating models for Computer Graphics, we added one major element to the basic simulator: a surface defined by an implicit function. Cells sense the gradient of the implicit function and move relative to it to find or avoid the surface. They also get an input describing the approximate feature size of the closest surface (so they can change to an appropriate size). We extended the simulator from 2-D to 3-D, but that did not involve any major architectural changes. One feature of the 2-D system that is not implemented in the 3-D system; the reactiondiffusion computation in the extracellular environment. We were concerned that the simple implementation with a discretized grid would be prohibitively expensive in 3D. From the viewpoint of Computer Graphics, the simulator is an enhanced version of a classic particle-system, where each cell is a particle. The main differences are: surface constraint, collision computation, availability of information about contact between particles, availability of orientation information from neighbors, and particle control via conditional differential equations