Compressing 3D Trees With Rendering Efficiency Based on Differential Data

Abstract

Trees or forests are indispensable and ubiquitous in virtual outdoor environments. Artists usually use high polygon counts to construct detailed 3D tree models to increase realism. However, large memory spaces are required, and considerable computation power is used for rendering. This paper proposes a compression method for 3D tree models to achieve rendering efficiency with less memory space and high visual fidelity of trees simultaneously. Trees are clustered using automatic K-medoids clustering method, and each member tree in the cluster can be reconstructed based on the representative tree with differential data. An abstract representation of an ordered rooted tree for each tree model is introduced, and the similarity between trees was measured using the modified tree edit distance. Moreover, an LOD priority for each component was associated to facilitate the LOD mechanism by considering the contribution to the visual perception after rendering. For accelerating rendering, GPU was exploited to benefit the LOD mechanism, and the geometry instancing facilitates the rendering for component instances shared among member trees. The effectiveness of compression was tested using four sample forests. As demonstrated by the experimental results, our method can save substantial memory space, retain the visual fidelity of the reconstructed trees, and accelerate the rendering.