針對蒙地卡羅路徑追蹤鏡面反射之路徑重複使用方法

Abstract

物理性質影像生成致力於產生逼真的影像，此技術被廣泛運用在許多產業之中，例如：電影產業中的電腦特效、建築與設計業中的3D模型預視；當然，還有電玩產業中的3D畫面。 光線的資訊在現實世界中是無限的，所以我們必須透過採樣的方式來進行估測。處理全域照明（Global Illumination）的方法，對於產生的影像是否看起來真實佔了很重要的腳色，不同的採樣方法會對場景的呈現產生不同的結果。雙向路徑追蹤（Bidirectional Path Tracing）是基於蒙地卡羅路徑追蹤（Monte Carlo Path Tracing）的許多方法之一，為一種對於場景運算可以快速收斂的方法；然而，因為先天上的限制，此方法無法處理某些特定狀況，例如：路徑有鏡面至粗糙面又再到鏡面（specular-diffuse-specular）的情況，會影響產出影像的品質。在此篇論文中，我們提出了一個基於雙向路徑追蹤方法，藉由路徑重複使用來處理此狀況的折衷對策。

Physically based image synthesis aims to render realistic images. It is widely used in many industries, e.g., Computer-generated imagery (CGI) in the movie industry, 3D models preview in the architecture and the design industry, and of course, 3D scenes in the computer gaming industry. In the real world, the information of light is infinite; therefore, we must do sampling to estimate it. How to deal with the global illumination effect plays an important role in whether the rendered image looks real or not. There are many sampling methods that reveal different features of the scene. Bidirectional path tracing, which is one of the Monte Carlo path tracing methods, is known for its quick convergence; however, due to innate limitation, it cannot handle some difficult cases, such as specular-diffuse-specular (SDS) paths. In this thesis, we introduce an alternative path reuse method while facing the SDS problem during the operation.