Author:

Abstract:

Photorealistic image synthesis techniques have matured in the past decades. Nowadays it is possible to create realistic computer-generated images on consumer hardware in minutes. In computer graphics research the focus has shifted to applying these well-known techniques on the graphics processing unit (GPU), which has evolved recently into a massively parallel general-purpose computing unit. This allows to generate realistic images interactively (multiple frames per second).This dissertation describes methods for creating photorealistic images interactively for scenes which contain glossy materials, such as shiny metals or plastics. The blurry reflections -typical for glossy materials- pose a significant challenge to known methods. The techniques proposed in this work provide solutions to this challenge and differ in terms of rendering speed, memory cost, preprocessing time and the quality of the resulting images.First of all we present a method for visualizing indirect highlights, which are formed by at least two glossy reflections between the camera and the light source. We show the difficulties from which known techniques suffer when computing such highlights and propose to compute a filtered response function to alleviate them. As a result we obtain faithful glossy reflections at near-interactive frame rates. Our method requires no preprocessing. We compare our results with ground-truth references and discuss possible sources of error.The remainder of this work discusses Precomputed Radiance Transfer (PRT) methods, which require preprocessing. PRT methods compute a transfer function on the surface of the scene. We demonstrate how to perform this computation adaptively, in order to focus calculations in regions where glossy reflections are to be expected. Our results run at interactive frame rates, at the expense of memory cost and preprocessing time.Finally we present a method to decrease both this preprocessing time and memory cost. The former by computing the transfer function iteratively, the latter by exploiting the sparsity of our transfer function representation. As a result we are able to generate photorealistic images at interactive frame rates of scenes with glossy materials. As such we are convinced that the work described in this dissertation is useful in the neverending quest for realism and speed in computer graphics.