预计算光能传输光照下逐像素渲染实现-realization of pixel-by-pixel rendering under pre-computed light energy transmission illumination.docxVIP

摘要摘要全局光照是计算机图像合成领域的一个永恒的话题。它着重于分析光的发 射、传播行为，并最终形成所谓的―照片真实‖的图像。其在CAD软件，三维计 算机游戏等领域有极其广泛的应用前景。但是，传统的全局光照生成算法所需 的存储量很大，计算复杂度很高，在图形处理单元（GPU）上难以实时实现。 因此，如何利用GPU的处理能力来实现漫反射和反光效果的全局光照，同时保 持低存储需求和高的帧率，是一个迫切需要解决的问题。本文利用预计算光能传输(PRT)技术，以球面谐波函数(SH)作为底层数据， 来实现在高动态范围环境下的逐像素全局光照效果。首先，根据特定光照条件， 利用球面谐波函数对场景信息进行预计算，将其编码为低存储需求的SH系数。 然后，在实时渲染阶段利用SH通过简单的数学运算来计算光照。另外，本文还 利用一种编码技术来降低存储和带宽的需求，而几乎不损失精度。通过延伸已 有的利用PRT在每顶点计算光照的技术，结合凹凸映射，可以用较少的存储量 和计算量，来实现全局光照效果。利用高动态范围（HDR）渲染，采用本文提出的方法可以将合成的场景实 时渲染到真实的环境中去。而且，对于很多实时应用而言，中等动态范围(MDR) 渲染就能获得相当不错的效果，可以满足交互或实时应用的需要。关键词：预计算光能传输，全局光照，实时渲染，球面谐波函数，高动态 范围渲染ABSTRACTABSTRACTGlobal illumination is an eternal topic of realistic computer image synthesis. It focuses on the behavior of light emission and transportation, and the so-called ―photo realistic‖ imagery that is derived from these fundamental principals. But the traditional way of doing global illumination is costly, both in terms of storage and pre-computation complexity, and thus does not suite well with the graphics processing unit(GPU) as far as real-time rendering is concerned. Therefore, it is a crucial problem to be solved that we implement real-time diffuse and glossy global illumination lighting with the GPU power, while maintaining a low storage cost and high frame rates level.This thesis utilizes Pre-computed Radiance Transfer(PRT) technique, and Spherical Harmonics(SH) functions as the data base, to implement global illumination lighting effects in a high dynamic range(HDR) environment. First, with specific lighting conditions, we pre-compute the scene with spherical harmonics functions and encode it as low storage cost SH coefficients. Then at real-time rendering stage we compute the lighting using Spherical Harmonics with simple mathematic calculations. In addition, we use an encoding technique to lower the storage and bandwidth requirements, while maintaining almost the same precision details. Extending existing per-vertex PRT lighting techniques, together with bump mapping, we implemen