Image-space approach to real-time realistic rendering

One of the main goals of computer graphics is the fast synthesis of photorealistic image of virtual 3D scenes. The work presented in this thesis addresses this goal of speed and realism. In real-time realistic rendering, we encounter certain problems that are difficult to solve in the traditional 3-dimensional geometric space. We show that using an image-space approach can provide effective solutions to these problems. Unlike geometric space algorithms that operate on 3D primitives such as points, edges, and polygons, image-space algorithms operate on 2D snapshot images of the 3D geometric data. Operating in image-space effectively decouples the geometric complexity of the 3D data from the run-time of the rendering algorithm. Other important advantages of image-space algorithms include ease of implementation on modern graphics hardware, and fast computation of approximate solutions to certain lighting calculations. We have applied the image-space approach and developed algorithms for three prominent problems in real-time realistic rendering, namely, representing and lighting large 3D scenes in the context of grass rendering, rendering caustics, which is a complex indirect illumination effect, and subsurface scattering for rendering of translucent objects.