Hierarchical annealing for synthesis of binary images

This paper introduces a hierarchical annealing algorithm which addresses the very large computational costs associated with simulated annealing for the synthesis of binary images. In real-world examples, the large configuration space of such models has led to the disappointing performance of annealing approaches. Our method essentially approaches different natural scales in the image separately and in a hierarchical manner. We demonstrate orders-of-magnitude improvement compared to existing results, and discuss the inherent computational difficulties encountered with these types of approaches.