Combining global and local global-illumination algorithms

Global illumination algorithms can be classified as local and global transfer methods. Local methods find a single point (or patch) in a given step and transfer its radiance towards other point(s). Global methods, on the other hand, select the source and the target of the transfer simultaneously. Local methods are better if the radiance distribution is heterogeneous and the scene is sparse, while global methods can win for dense scenes of homogeneous radiance. This paper proposes the combination of global and local global illumination algorithms in the sense of multiple importance sampling. In this way, the combined method can eliminate the higher noise at the corners produced by local methods and the need for first-shot for global techniques.