It just fills in the gaps with a whole lot of imagination, not too different from how a human would. Unless it has access to a more detailed picture or contextual information, it cannot extract more information than was actually captured.