Imaging for dismantlement verification: information management and analysis algorithms
The level of detail discernible in imaging techniques has generally excluded them from consideration as verification tools in inspection regimes. An image will almost certainly contain highly sensitive information, and storing a comparison image will almost certainly violate a cardinal principle of information barriers: that no sensitive information be stored in the system. To overcome this problem, some features of the image might be reduced to a few parameters suitable for definition as an attribute, which must be non-sensitive to be acceptable in an Information Barrier regime. However, this process must be performed with care. Features like the perimeter, area, and intensity of an object, for example, might reveal sensitive information. Any data-reduction technique must provide sufficient information to discriminate a real object from a spoofed or incorrect one, while avoiding disclosure (or storage) of any sensitive object qualities. Ultimately, algorithms are intended to provide only a yes/no response verifying the presence of features in the image. We discuss the utility of imaging for arms control applications and present three image-based verification algorithms in this context. The algorithms reduce full image information to non-sensitive feature information, in a process that is intended to enable verification while eliminating the possibility of image reconstruction. The underlying images can be highly detailed, since they are dynamically generated behind an information barrier. We consider the use of active (conventional) radiography alone and in tandem with passive (auto) radiography. We study these algorithms in terms of technical performance in image analysis and application to an information barrier scheme.