样本驱动的大规模植被分布高效合成

A High-Efficient Algorithm for Sample-Based Synthesis of Vegetation Distribution

针对自然场景中丰富多样的植被分布模式的模拟时,现有的通用拟合算法计算耗时,不能满足大规模分布与实时交互分布的需要.为了解决任意分布模式的快速合成问题,提出一种基于样本分布控制并结合Wang tiles优化的高效点分布合成方法.该方法由离线预计算过程和实时拼接过程组成.在离线预计算过程中,首先通过基于环形点对距离统计的对相关函数特征拟合过程合成多块首尾分布连续,以及与样本分布相似的初始点分布tile,然后通过对初始tile的边界处理得到一组可以非周期性拼接且接缝处分布保持连续的Wang tiles;在此基础上,利用这组Wang tiles进行实时拼接,从而实现以极快的速度和极小的内存占用合成大规模的分布的目的.实验结果表明,该方法适用于随机、均匀、聚集等各种分布特征的数据,是一种对各类样本分布模式通用的分布合成方法;并在大规模分布的合成上比原有的基于样本驱动的分布拟合算法达到3～6个量级的速度提升;可有效地应用于自然场景模拟中大规模植被的分布信息构建,及其他有实时交互分布需求的应用中.

For the simulation of diverse distribution patterns of vegetation in natural scenes,the existing general distribution fitting algorithm was too time-consuming to meet the needs of large-scale distribution or real-time interactive distribution synthesis.In order to achieve rapid synthesis of arbitrary distribution patterns,an efficient point distribution synthesis method based on sample patterns control combined with Wang tiles acceleration is proposed.The method consists of two key phases:an off-line pre-computing phase and a real-time tiling phase.In the pre-computing phase,the PCF feature fitting process based on toroidal point pair distance statistics is used to synthesize initial point distribution tiles with similar distribution patterns to samples and end to end feature continuity.Then a set of Wang tiles that can be tiled non-periodically and seamlessly is obtained by adjusting the boundaries of the initial tile.With these Wang tiles,the real-time tiling phase can synthesize large-scale distribution at extremely fast speed and with very small memory footprint.This method is a general distribution synthesis method applicable to data with various distribution patterns(random,uniform,and clumped),and achieves a speed improvement of 3-6 orders of magnitude over the original sample-based algorithm in the large-scale distribution synthesis.This method can be effectively applied to the construction of large-scale vegetation scenes and other applications with real-time interactive distribution.