Hierarchical annealing for random image synthesis

Simulated annealing has been applied to a wide variety of problems in image processing and synthesis. However, particularly in scientific applications, the computational complexity of annealing may constrain its effectiveness, in that the demand for very high resolution samples or even three-dimensional data may result in huge configuration spaces. In this paper a method of hierarchical simulated annealing is introduced, which can lead to large gains in computational complexity for suitable models. As an example, the approach is applied to the synthesis of binary porous media images.