In this work, we present a method of controlling self-assembly. First, we demonstrate that images of surface morphologies can be decomposed into important features that capture essential properties of the surface. Subsequently, we develop a method for obtaining a low-dimensional representation of the entire feature set and show that this reduced representation can be used as a set of surrogate process outputs for process control and design of thin-films. Finally, the effectiveness of this approach is demonstrated in several simulation studies that utilize two macroscopic manipulated variables known to influence the self-assembly of thin-films, temperature and repulsion strength, to drive arbitrary patterns toward a target morphology. (C) 2014 Elsevier Ltd. All rights reserved.