The information comes from a set of slides from 2011 found in Edward Snowden's trove of NSA documents.
The images are harvested from the various communications the spy agency intercepts via its surveillance operations around the world (from emails, text messages, videoconferences), as well as from the internet (from social media, commercial facial databases, and other web resources).
They are also collected from other databases such as national identity card databases set up by foreign countries - whether they are permitted to access them or not.
The NSA combines the images with biometric, biographic and contextual information it has on individuals.
"Agency officials believe that technological advances could revolutionize the way that the NSA finds intelligence targets around the world, the documents show. The agency’s ambitions for this highly sensitive ability and the scale of its effort have not previously been disclosed," the reporters noted.
"While once focused on written and oral communications, the NSA now considers facial images, fingerprints and other identifiers just as important to its mission of tracking suspected terrorists and other intelligence targets."
The facial recognition technology they used is powerful and flawed at the same time: sometimes is does zero-in on the suspect immediately, other times it gives many false positives, some of which would be immediately obvious to humans. It narrows the haystack, for sure, but it also occasionally leads to errors.
NSA's spokeswoman has commented on the disclosures by saying that the agency does not mine photographs in American states' databases of driver’s licenses or passport, but has not disclosed whether the NSA has access to the State Department's database containing the photos of foreigners who apply for a US visa.