An Image-Based Radiosity Representation

Abstract

We present a geometry-independent radiosity representation which stems from image-based methodologies. Our image-based representation is inspired by the concept of layered depth images. To construct the radiosity function we render shaded depth images and warp them into the representation. This enables us to alter Keller’s instant radiosity algorithm to a view-independent variant, allowing interactive viewing of complex scenes without relighting every frame. Image warping is done on the host CPU, while graphics hardware is exploited for rendering the images. Our approach differs from the traditional finite element methods since no a priori meshing, nor an explicit form factor evaluation needs to be performed. The geometry independence ensures a reliable rendering time and memory usage, which are more dependent on image resolution rather than scene complexity. It also implies that lighting information is decoupled from scene representation, thereby avoiding the restriction to planar surfaces and issues such as discontinuity meshing.