Learning text to model: A Bayesian network based L-system modeling strategy

Abstract

L-system is a prevailing modeling method for generating fractals, especially self-similar patterns such as plants. However it's too hard to design an appropriate L-system to get the desired visual models of plants. In order to generate a favorable plant model, usually we need to deduce backwards or guess the production rules of the L-system and then try to modify some control parameters over and over again. Inspired by information extraction technology, we propose a new strategy to model visual plants. We use Bayesian Networks to extract structured information describing the plant characters from user given text first, then we use that information to automatically generate an L-system alphabet, axiom and production rules. Comprehensive experimental evaluation conducted on real botanic text corpora demonstrates that our proposal is very helpful in artistic plants modelling.