What they're doing is a high-resolution fMRI scan of V1 (primary visual cortex), a region of the brain that shows a "retinotopic map" of the visual field. It's been known for a while that it's possible to reconstruct an image from V1 activation. What this research shows is that fMRI imaging is now detailed enough to do this non-invasively.
Also, something cool about this research, and also a bit scary, is that it's also known that this region is activated by visual recall (a quick search revealed this: http://www.pubmedcentral.nih.gov/articlerender.fcgi?artid=48...). I'm not sure if anyone has ever tried to reproduce an image from V1 activation during visual recall, so I don't know if it would be a coherent image. My guess is it would though, in which case an fMRI scan could detect an image of an "N" if you asked the person to picture the letter.
Also, something cool about this research, and also a bit scary, is that it's also known that this region is activated by visual recall (a quick search revealed this: http://www.pubmedcentral.nih.gov/articlerender.fcgi?artid=48...). I'm not sure if anyone has ever tried to reproduce an image from V1 activation during visual recall, so I don't know if it would be a coherent image. My guess is it would though, in which case an fMRI scan could detect an image of an "N" if you asked the person to picture the letter.
Pretty cool.