I'm not sure how anyone would have any capability to 'reverse' the final product. Such processing is 'impossible', if I might be so bold. Pixel degradation alone, one might only be able to 'hypersample' to a very nominal extent to give the perception of increased resolution, but either way you're destroying and corrupting the data to become only useful in limited case scenarios. 'Hypersampling' might only be useful as a vis-sim product, for example -- though in limited use scenarios, especially if equal resolution classification is being required as a component product by the application.

I do understand the hang-ups in the bureaucracy involved with 'concerns' NGA might have while they flushing such ideas out of their system, crossing i's, dotting t's, however, and such and whatnot.