暫存資訊在即時屏幕空間全域照明技術之應用

Abstract

模擬全域照明現象在電腦圖學的領域中，一直是火熱的研究議題。漫射相互反射、焦散、折射和軟陰影等現象提供了重要的視覺線索，並且它們的加入可大幅地改善圖像的真實感，而為了做更多的即時應用，像是電腦遊戲或虛擬實境系統，能即時的計算全域照明現象是很重要的。 本篇論文結合反射陰影圖(Reflective Shadow Maps)、螢幕空間環境遮蔽(Screen Space Ambient Occlusion)及百分比漸進濾波(Percentage Closer Filtering)等方式繪製出逼真的場景，並藉由改變取樣模式和累積前一張畫面的資訊提高圖像品質，最後利用反向重投影法(Reverse Re-projection)使得收斂好的畫面在鏡頭移動的過程中盡可能地重複利用，達到效能提升，經由實作證明本演算法在大型的場景下也有不錯的表現。

Simulating global illumination has been the intense research topic in computer graphics for many years. Effects such as diffuse inter-reflection, caustics, refractions and soft shadows provide the important visual cues that help the viewer to understand spatial relationships and also drastically improve the realism of a rendered scene. Thus, in order to do many real-time applications like computer games and virtual reality systems, computing global illumination at interactive frame rates is significant.

In this thesis, we propose an algorithm combining reflective shadow maps, screen space ambient occlusion and percentage closer filtering for realistic image synthesis. By changing sampling patterns and accumulating the results of previous frames, we can improve the quality. Finally, in order to achieve high performance, we exploit reverse re-projection technique to reuse converged information while the camera is moving. We demonstrate that the algorithm works well for large-scale scenes at interactive frame rates.