Inverse procedural modeling of trees

Ondrej Stava, Purdue University

Abstract

Despite the number of new tree-modeling techniques that have been proposed over the past thirty years, the internal representations of trees have remained mostly unchanged. Trees are usually represented either explicitly by their geometry or implicitly as a procedural model. Unfortunately, both of these representations have many inherent limitations that in practice complicate the application of existing tree-modeling methods. The geometric representations of trees are static, and they cannot be reused in different environmental settings or to form large ecosystems. Reusability is the typical property of procedural representations of trees, but such representations are notoriously difficult to design, and they do not provide a sufficient level of direct control to meet specific artistic requirements. To combine advantages of both representations, the geometric and the procedural, it is necessary to have a mechanism that would enable us to convert one representation to the other at any point during the modeling process. Although it is easy to convert a procedural representation into a geometric one, the inverse process is much more complicated. To address this problem, we propose a new inverse procedural approach that automatically creates a procedural model from a given input geometric representation of a tree. We evaluate our method on a set of different tree models that represent a broad range of tree species originating from various sources, including scans of real-world trees, tree libraries, and trees generated by known procedural models.

Degree

Ph.D.

Advisors

Benes, Purdue University.

Subject Area

Plant biology|Computer science

COinS