Efficient computations of scalable caustic rendering and reconstruction
Date of Issue2015
School of Computer Engineering
In Computer Graphics, research on caustic can be divided into two categories: caustics rendering and inverse caustics. In general, caustic computations are expensive. Therefore, one of the main challenges in caustic computation is on the scalability issue. Firstly, it is challenging to render real-time caustics of scenes under environment illumination as we must consider all light from environment. Secondly, it requires a very high computational cost to render spectral caustics (rainbow-like colour effects) since we must consider the whole range of visible wavelengths. Lastly, in the inverse caustic computation, i.e. computing the geometry of a caustic object (an object that reﬂects and/or refracts light) given an input caustic pattern, most work focus only on a single input caustic pattern. For multiple caustic patterns, it is a challenging problem due to the shape and size differences among the input caustic patterns which are difﬁcult to satisfy. In this thesis, we address these scalability issues in caustic research. Our main idea in solving these issues is adjustment of inputs such that the problems are tractable. We present our solutions as follows. Real-time caustics under environment illumination In this research, we compute real-time caustics of a rigid caustic object under environment illumination (represented as an environment cube map). To achieve this, we precompute the caustic patterns on the surrounding space of a caustic object based on several predeﬁned directional lights. In the rendering, we adjust the environment light source by approximating it as several directional lights by using our proposed environment cube map segmentation technique and integrate their contributions in the rendering. Our technique is able to render cast and volumetric caustics under environment illumination in real-time by using GPU. Spectral caustic rendering We propose a two-step acceleration scheme by considering spectral characteristics of the scene elements, such as index of refraction distribution of caustic objects and spectral reﬂectance of surrounding surfaces. In our ﬁrst acceleration step, instead of spawning rays for every visible wavelength, we adjust the visible wavelength inputs by clustering them based on the similarity of the refraction angle such that we can represent several wavelengths as one light ray. In the second acceleration step, we compute a scene-dependant importance level of each wavelength cluster in order to determine the reﬁnement iteration amount. Our scheme can achieve speed up of up to around 100 times while maintaining the rendering quality. Inverse caustics We compute the geometry of a caustic object that can reconstruct a set of input caustic patterns with each caustic pattern is located at an user-input distance from the caustic object. The inputs pose difﬁcult constraints due to the differences in the caustic patterns to be satisﬁed. To solve this problem, we propose a two-step optimization technique in which we adjust the position and size of the caustic regions in the ﬁrst step and we adjust the caustic shapes in the second step. We also present an additional step to improve the intensity of caustic patterns. Our technique is able to construct a caustic object for various types of input patterns and we validate the results by using mental ray rendering. Our work has several applications, such as accelerated rendering for various multimedia applications (e.g. computer games and simulations), and arts (computer-generated movies and pictures, and physical installation of caustic objects).
DRNTU::Engineering::Computer science and engineering::Computing methodologies::Computer graphics