多层物理材质表达和高效绘制方法研究

Realism becomes more and more significant in movie production, game development and virtual reality fields. It requires both physically based complex material representation and efficient realistic rendering methods. The current material models usually ignore that the materials have multiple layers, and this leads to unrealistic materials, such as wood ground. Furthermore, the current material models are based on geometrical optical models, thus it cannot support Nano features, such as diffraction. From another side, a large amount of light transport simulation is required in order to keep the complex materials realistic, such as participating media. It costs long time and large memory to simulate. To solve unrealistic material and inefficient rendering issues, we propose physically based material representation and efficient rendering methods. Firstly, we propose a multiple layered model for two-scale microfacet to represent the complex materials more accurately. Secondly, we propose a precomputed multiple scattering model in Monte Carlo based methods and a point-based method to simulate the light transport of participating media under complex scenes to obtain efficient participating media rendering. Finally, we propose a convolutional neural network based Monte Carlo denoising method suitable for complex materials and participating media. Our project will enrich theories and techniques of physically based material representation and efficient participating media rendering. The proposed techniques can also be applied in movie production, game development, virtual reality fields, etc.

影视制作、游戏开发以及虚拟现实等领域对真实感绘制的要求与日俱增，这既需要满足物理规律的复杂材质表达模型，又需要高效的真实感绘制方法。目前基于物理的材质模型忽略材质多层性，使得地板等材质缺乏真实感；仅限于几何光学模型，难以呈现纳米级特性，比如光的衍射。为增强真实感，需要模拟大量光线传递过程，才能保证参与介质等复杂材质的绘制效果，这给时间和存储均带来挑战。本项目针对材质缺乏真实感、绘制效率低等问题，提出遵循物理规律的材质表达机理和高效高质绘制方法，包括：研究多层双尺度微表面材质模型，更加精准地表示现实世界中的材质；研究基于多次散射模型预计算模型的蒙特卡洛方法和针对复杂材质的点云缓存绘制算法，实现参与介质的高效绘制；研究基于卷积神经网络的复杂物理材质绘制降噪方法，进一步提升渲染质量。本项目的研究成果将为物理材质表达和高效绘制提供理论和技术支持，应用到动画制作、游戏设计以及虚拟现实等领域中。

影视制作、游戏开发以及虚拟现实等领域对真实感绘制的要求与日俱增，这既需要满足物理规律的复杂材质表达模型，又需要高效的真实感绘制方法。目前基于物理的材质模型在微结构模型方面存在效率低、合成困难等问题。为增强真实感，需要模拟大量光线传递过程，才能保证参与介质等复杂材质的绘制效果，这给时间和存储均带来挑战。本项目针对材质缺乏真实感、绘制效率低等问题，提出遵循物理规律的材质表达机理和高效高质绘制方法，包括：研究微结构材质模型，更加精准地表示现实世界中的材质；研究基于多次散射模型预计算模型的蒙特卡洛方法和针对复杂材质的点云缓存绘制算法，实现参与介质的高效绘制；研究基于卷积神经网络的复杂物理材质绘制降噪方法，进一步提升渲染质量。本项目的研究成果将为物理材质表达和高效绘制提供理论和技术支持，应用到动画制作、游戏设计以及虚拟现实等领域中。