用於球諧函數照明架構之有效取樣與重建入射光強度之方法

Abstract

球諧函數照明為一即時顯像技術，其能夠有效率的顯示物體在受到動態、變化緩慢的遠方光源照明之下的陰影等光影效果。為了讓此架構可以處理由鄰近光源所造成的照明效果，我們提出一個取樣與重建的架構。在之前的方法中，鄰近光源所形成的入射光強度函數，是透過一個前處理程序來均勻散佈取樣位置。這些樣本接著用一個跟距離有關的函數進行內插，以得到物體表面每一個頂點的入射光強度。由於樣本的散佈方式並不隨著入射光強度而變化，加上內插時所有的樣本不論遠近均被考慮進來，入射光強度的局部變化無法被成功重建。我們提出一個動態根據物體幾何以及入射光強度來散佈樣本的方法，這個方法可以讓使用者指定一錯誤容忍值，以便在樣本數量與重建品質之間進行控制。另外，對於每個樣本，我們都給予一個有效範圍，使得入射光強度的局部變化能夠被有效保存，並且提升重建效率。

Pre-computed radiance transfer accounting for self-shadowing and interreflection allows objects to be shaded by distant, dynamic, low-frequency lights [SKS02]. In the previous work, incident radiance function is sampled at points uniformly selected by an off-line process, and a radial function is then used to interpolate all samples. If lights are near to objects, or occluding one another, a large number of samples are required. Since interpolation is performed through all samples, locality can be likely to lose. In this thesis, we present a framework of sampling and reconstruction in which objects can be shaded by nearby luminaries. Our work dynamically selects sample points according to object geometries and nearby luminaries. The quality of sampling is controlled by a user-specified error bound. Furthermore, the valid domain of samples can be used as error bounds to enhance the interpolation.