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 will 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 classified data security while maintaining high confidence for verification. The proposed work is focused on developing secure, robust, tamper-sensitive and automatic techniques that may enable the comparison of non-sensitive hashed image data outside an IB. It is rooted in research on so-called perceptual hash functions for image comparison, at the interface of signal/image processing, pattern recognition, cryptography, and information theory. Such perceptual or robust image hashing—which, strictly speaking, is not truly cryptographic hashing—has extensive application in content authentication and information retrieval, database search, and security assurance. Applying and extending the principles of perceptual hashing to imaging for arms control, we propose techniques that are sensitive to altering, forging and tampering of the imaged object yet robust and tolerant to content-preserving image distortions and noise. Ensuring that the information contained in the hashed image data (available out-of-IB) cannot be used to extract sensitive information about the imaged object is of primary concern. Thus the techniques are characterized by high unpredictability to guarantee security. We will present an assessment of the performance of our techniques with respect to security, sensitivity and robustness on the basis of a methodical and mathematically precise framework.
- Research Organization:
- Pacific Northwest National Lab. (PNNL), Richland, WA (United States)
- Sponsoring Organization:
- USDOE
- DOE Contract Number:
- AC05-76RL01830
- OSTI ID:
- 1149246
- Report Number(s):
- PNNL-SA-88946; NN2001000
- Resource Relation:
- Conference: 53rd Annual Meeting of the Institute for Nuclear Materials Management (INMM2012), July 15-19, 2012, Orlando, Florida , 2:1133-1142
- Country of Publication:
- United States
- Language:
- English
Similar Records
Low-Intrusion Techniques and Sensitive Information Management for Warhead Counting and Verification: FY2011 Annual Report
Concept of Operations for Nuclear Warhead Embedded Sensors