Simulated annealing and relaxation models are not new. Hierarchical models of visual processing are not new. What is new in the present model is the combination of these two concepts by way of the reciprocal feedback through forward and inverse transformations.
As to the relaxation algorithms to which the reviewer refers, which deal with Gestalt issues, the reviewer leaves it to our imagination as to which specific papers he is referring to. Would any of those relaxation algorithms happen to involve forward and inverse transformations across levels in a hierarchical representation?