貼圖空間重要取樣之即時半透明顯像技術

Abstract

我們提出了一個利用貼圖空間的即時半透明材質顯像方法，貼圖空間的取得來自於模型表面參數化(surface parameterization)。我們將原先在三維模型表面上的積分轉換成一個在二維貼圖空間的積分式，並利用了以進入幅射量(irradiance)為基礎的重要取樣(importance sampling)來計算二維貼圖空間的積分。我們的架構可以在動態光源、動態材質以及即時的情況下利用少量的前置運算資料(precomputation overhead)產生正確的半透明材質影像。當半透明現象明顯地由區域效果(local effects)所主導時，我們方法會受困於取樣數量需要相當大的問題。為了克服這問題，我們提出了混合式架構。在這架構裡，我們展示半透明材質顯像如何能被分解成區域效果(local effects)以及全域效果(global effects)。我們應用我們的貼圖空間重點取樣方法在全域效果的計算上。對於區域效果的計算則是結合兩個現有的即時半透明顯像方法[2,15]。我們的混合式架構在動態的材質下，可以用穩定效能與取樣數量來顯像半透明材質。

We present a novel approach for real-time translucent rendering by using the texture space of the model surface, which is achieved from surface parameterization. We convert the integration over a 3D model surface into an integration over a 2D texture space. We apply importance sampling based on the irradiance for the integration over the texture space. We propose a threestage GPU-based rendering approach for translucent rendering. Our method can render accurate translucent image in real-time under dynamic environment about light and materials with lower precomputation overhead. Our method suffers from problem of sample number when the appearance of translucency is dominant to local effect. We propose hybrid method to overcome this problem. In our hybrid method, we show how the translucent rendering can be decomposed as local effects and global effects. We apply our GPU-based texture space importance sampling approach on the evaluation of global effects. The local effects are evaluated by the combination of two available methods for real-time translucent rendering [2, 15]. Our hybrid method can steadily render dynamic translucent material without changing the number of samples.