Research On Annotation-learning-based Diversified Tree Modeling

Virtual reality technology has always been a hotspot in computer graphics. Asan efficient way to represent objects in virtual world, a rapid reconstruction ofthree-dimensional(3D) model is gaining extensive attention. Being a commonphenomenon, the diversity of trees determines the complexity of theircorresponding3D models, and the realism of tree models has an strong influence onthe immersion of virtual view. As a result, the simulation of tree growth and themodeling of landscape have been the focus of study. Though automatic modelingtechnique achieves quite good effect, user can only define effective generationrules with specialized knowledge of trees, and it is difficult for user to understandthe overall structure because of the abstract modeling process. Interaction-basedmethod can control the tree modeling process conveniently, but the result is greatlyaffected by human experience, and user has to repeat the processes for each tree.Therefore, how to model mass trees with different structural features by using aunified technique is an important problem.Based on inverse procedural modeling, the rapid automatic3D tree modelingmethod has been studied in this dissertation by introducing learning mechanism andtaking user-identified information as guidance, and it mainly includes the followingpoints:(1) Key information which influences tree appearance is defined based on aninvestigation of branches in botany; annotations are added to the treestructure according to these key information; a Bayes Network feature databaseof trees is established by extracting features from a plenty of pictures.(2) Production rules of L-systems are defined according to the branch characters oftrees in botany after studying modeling methods based on L-systems; the BayesNetwork is used to conduct the modeling process under the guide of featuredatabase.(3) Leaves modeling of different trees are realized by using the transparent textureand leaves layout algorithm after exploring the modeling technique for singleleaf; this modeling method is integrated to the annotation-learning-baseddiversified tree modeling system.