Information Barriers for Imaging: Scale-Invariant Feature Transformations and Homomorphic Encryption

Imaging techniques are of growing interest for future arms control treaties, in light of their unique ability to identify form and function of a declared treaty-limited item. Yet it is precisely the level of revealed detail in form and function that has traditionally meant a reluctance to consider imaging for this purpose. With an ultimate goal of demonstrated security of imaging techniques, several current efforts aim to incorporate information protection directly into image data collection and/or processing. We discuss a demonstration of one such approach that we have developed to combine a feature vector as a computer hash of an image with enhanced information protection via homomorphic encryption. In this approach, a Scale-Invariant Feature Transformation (SIFT) template is formed from a trusted treaty-limited item radiograph, and a corresponding SIFT of a test item radiograph is compared to the template to confirm (or reject) the tested item. In principle, the feature-vector comparison would occur within homomorphic encryption to eliminate the need for more complicated software information barriers. We present results of a demonstration of the previously developed SIFT framework. Although SIFT results theoretically obscure any actual image detail, SIFT alone is not sufficient in that regard. We conclude with latest results on the possibility of implementing homomorphic encryption to close the information protection loop.