We’ve all winced at those scenes in CSI:Nowheresville when somebody in a lab coat ‘enhances’ an image until you can see the reflection of the room in the pupil of somebody’s eye, or whatever.
This video is like that. Clearly implausible, most likely witchcraft.
Alternatively, a group of researchers at MIT really have found a way of extracting temporal information from CMOS sensor skew, sufficient to reconstruct audio up to around the 400Hz range from pictures alone.
Damned impressive, even if ‘sorcery’ is a more plausible algorithm.