一個事先計算輻射傳輸為基礎的即時全域照度演算法

Abstract

在全域照明當中，我們可以使用事先計算輻射傳輸函數，來儲存光線照明的果。由於事先計算輻射傳輸函數的資料量龐大，光線照明的效果並不能即時的描繪出來。在過去的研究中，利用分類的概念來事先計算輻射傳輸函數資料的壓縮，來加速描繪光線照明效果的效率。但是資料分類會導致了多邊形重繪的問題。多邊形重繪會直接的降低描繪的效率。在本論文當中，我們利用了位置一致性來減少多邊形重繪。另外我們也提出了一個選擇分類為超分類的演算法來減少多邊形重繪。

In global illumination, we can use Pre-computed Radiance Transfer (PRT) Function to store the lighting effects. Since the required storage for the values of PRT Function is huge, the result of lighting can not be rendered in real-time. In Peter-Pike Sloan‘s research, they compress the required storage by clustering PRT data to accelerate the rendering performance. But clustering data may result the polygon overdraw problem. Polygon overdraw will decrease the rendering performance directly. In this thesis, we apply the Geometry Coherence to reduce overdraw. Besides, we also introduce a cluster selection algorithm for super-cluster to decrease triangle overdraws.