I wouldn't be surprised if QA is completely automated right about the time that facial recognition becomes viable. I guess that means many years from now, but if the hitch to automating QA is having person-eyes looking at a page and using the site in order to find problems, then I think they might ultimately converge their tooling because there's similarities in there. Something you can throw your PSDs or Balsamiq URLs at it and wait for the notifications. Web page recognition rather than facial. Heck, it might be simpler...or already exist.

Update: Here's a guy doing text recognition for localization: http://www.jebriggs.com/blog/2013/04/imug-computer-vision-an...

Also: http://browserbite.com/

Comparing screenshots from Selenium is a pretty nice way of doing it.