Theory and Algorithms for Efficient Physically-based Illumination

Author Jaakko Lehtinen Title Theory and Algorithms for Efficient Physically-Based Illumination Realistic image synthesis is one of the central fields of study within computer graphics. This thesis treats efficient methods for simulating light transport in situations where the incident illumination is produced by non-pointlike area light sources and distant illumination described by environment maps. We describe novel theory and algorithms for physicallybased lighting computations, and expose the design choices and tradeoffs on which the techniques are based. Two publications included in this thesis deal with precomputed light transport. These techniques produce interactive renderings of static scenes under dynamic illumination and full global illumination effects. This is achieved through sacrificing the ability to freely deform and move the objects in the scene. We present a comprehensive mathematical framework for precomputed light transport. The framework, which is given as an abstract operator equation that extends the well-known rendering equation, encompasses a significant amount of prior work as its special cases. We also present a particular method for rendering objects in low-frequency lighting environments, where increased efficiency is gained through the use of compactly supported function bases. Physically-based shadows from area and environmental light sources are an important factor in perceived image realism. We present two algorithms for shadow computation. The first technique computes shadows cast by low-frequency environmental illumination on animated objects at interactive rates without requiring difficult precomputation or a priori knowledge of the animations. Here the capability to animate is gained by forfeiting indirect illumination. Another novel shadow algorithm for off-line rendering significantly enhances a previous physically-based soft shadow technique by introducing an improved spatial hierarchy that alleviates redundant computations at the cost of using more memory. This thesis advances the state of the art in realistic image synthesis by introducing several algorithms that are more efficient than their predecessors. Furthermore, the theoretical contributions should enable the transfer of ideas from one particular application to others through abstract generalization of the underlying mathematical concepts. UDC 004.925, 004.925.3