Image Hashes as Templates for Verification

Imaging systems can provide measurements that confidently assess characteristics of nuclear weapons and dismantled weapon components, and such assessment may be needed in future verification for arms control. Yet imaging is often viewed as too intrusive, raising concern about the ability to protect sensitive information. In particular, the prospect of using image-based templates for verifying the presence or absence of a warhead, or of the declared configuration of fissile material in storage, may be rejected out-of-hand as being too vulnerable to violation of information barrier (IB) principles. Development of a rigorous approach for generating and comparing reduced-information templates from images, and assessing the security, sensitivity, and robustness of verification using such templates, are needed to address these concerns. We discuss our efforts to develop such a rigorous approach based on a combination of image-feature extraction and encryption-utilizing hash functions to confirm proffered declarations, providing strong sensitive data security while maintaining high confidence for verification. The proposed work is focused on developing automated techniques that may enable the comparison of non-sensitive hashed image data outside an IB. We present an assessment of the performance of our techniques on the basis of a methodical and mathematically precise framework.