利用能見度近似之反射性陰影圖改良

Abstract

反射性陰影圖是個被廣泛應用於即時影像生成的全域照明演算法，它延伸了傳統陰影圖之概念與應用，並且能繪製經過第一次反射的間接照明效果。雖然反射性陰影圖被視為一個效率良好的演算法，然而出自於運算複雜度方面的考量，它並不針對間接照明過程中物體對光線之遮蔽效應進行處理。在實際應用當中，若是忽略間接光源在照明過程中被場景物體所遮蔽之效應，經常會造成諸如照明效果不正確地溢出等的負面結果。為了解決此問題，我們以反射性陰影圖做為基礎，進而提出一個有效率的能見度估計演算法以及一個全新的採樣模式。所有原本演算法的優點都被我們刻意保留，如：完全運作於螢幕空間內、不須對資料進行任何預先處理等，因此我們的演算法同樣適合於高度動態之場景與照明環境。另外，我們尚詳述了針對當代圖形處理單元（GPU）所設計的演算法實作系統，其中引入了許多常用的即時繪圖加速技巧以達到高度效能表現。最後的實驗結果顯示，我們所提出之改良演算法相較於傳統反射性陰影圖，能夠產生更為真實的影像。

In this work we propose an improved Reflective Shadow Map algorithm by approximating visibility to account for indirect shadows. Reflective Shadow Map (RSM) is a popular approach for real-time global illumination. It approximates the first-bounce indirect lighting by using an extension of standard shadow maps. RSM, however, does not take visibility into consideration. Ignoring occlusion between secondary light source and shading point would result in light leakage and produce undesired results. To address this problem, we present an efficient algorithm for visibility approximation and a novel sampling pattern that cooperates well with it. Our approach works entirely in screen space and requires no precomputation of any sort. Therefore, it is suitable for highly dynamic scenes and lighting changing environments. We also describe a detailed implementation on contemporary GPU with many widely used real-time rendering techniques to achieve high performance. Results show the image quality produced by our approach is more realistic compared to the original RSM.