双向光割技术在多光绘制中的应用研究

发布时间：2018-07-22 12:18

【摘要】：多光方法在解决全局光照问题中提供了一个非常通用的框架,使得我们可以使用数十万以上的虚拟点光源来绘制出多种不同的特效。在一个很大的虚拟点光源集合中,每个虚拟点光源的贡献其实不是相等的。一般地,可扩展的算法都试图利用这种分布不均匀的虚拟点光源重要性来减少计算量,旨在识别和评估所有重要的虚拟点光源,另一方面对于不重要的虚拟点光源则只进行稀疏的估值。为了高效地收集这些虚拟点光源的贡献估值,光割算法及其扩展方法将这些虚拟点光源进行了聚类,从而使用代表类来近似表示多光绘制矩阵。而双向光割算法在此基础上采用双向框架以及更深的视点子路径来扩展支持更多的基于虚拟点光源绘制的材质。本文在Mitsuba系统上实现了双向光割算法,能够生成出高光泽、离线的高质量图像,能够比较好的兼顾程序运行效率和绘制效果。首先,我们执行通用的多光方法初始化模块生成计算所需要的数据。接下来,对于光割积分器,我们使用光树这一数据结构对生成好的虚拟点光源进行聚类；对于多维光割积分器,则还需要对虚拟感知器进行聚类；双向光割积分器则在上述操作之前增加了路径采样,以实现对虚拟感知器节点路径的扩展,并记录相关的信息。最后,我们利用光树和虚拟感知器节点树对光割或者多维光割进行划分,从而得到绘制结果。在文章的最后,我们展示了所实现积分器的绘制效果,并进行了相关的实验分析。
[Abstract]:The multi-light method provides a very general framework for solving the global illumination problem, which enables us to use hundreds of thousands of virtual point light sources to draw various special effects. In a large set of virtual point light sources, the contribution of each virtual point light source is not equal. In general, scalable algorithms attempt to reduce computational complexity by exploiting the importance of this uneven distribution of virtual point light sources to identify and evaluate all important virtual point light sources. On the other hand, only sparse estimates are made for unimportant virtual point light sources. In order to efficiently collect the contribution estimates of these virtual point light sources, the optical cutting algorithm and its extension method cluster these virtual point light sources, and use representative classes to approximate represent the multi-light rendering matrix. On the basis of this, bidirectional optical cutting algorithm uses bidirectional frame and deeper view sub-path to extend the material which can be drawn based on virtual point light source. In this paper, a bidirectional optical cutting algorithm is implemented on Mitsubishi system, which can produce high gloss and high quality off-line images, and can give good consideration to the efficiency of program operation and the effect of drawing. First, we perform a general multi-optical initialization module to generate the required data. Then, we use the data structure of optical tree to cluster the generated virtual point light source, and for the multidimensional optical cutting integrator, we also need to cluster the virtual perceptron. The bidirectional optical cutting integrator adds the path sampling before the above operation to realize the expansion of the path of the virtual perceptron node and record the related information. Finally, we use the optical tree and the virtual perceptron node tree to divide the optical or multi-dimensional light cuts, and obtain the rendering results. At the end of the paper, we show the rendering effect of the integrator and analyze the experiment.
【学位授予单位】：浙江大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TP391.41

【相似文献】