Buried in a post on search quality on the Google blog was an interesting titbit:
we now offer an early form of face recognition on advanced search
Frustratingly that was all there was – it would be interesting to hear if this is technology they had developed themselves or if they are using a third party such as Viewdle, Like.com or even Blinkx, each of whom are innovating in this area.
Visual search is one of the great challenges in unlocking the potential metadata hidden in images and video. Alt text, image name and descriptions only take you so far in understanding a video or image’s content. They may do a good job of describing the overall content, e.g. visit to Paris, but struggle to explain either specific components of the static image (Eiffel Tower in background) or content variations within a longer piece of video (Notre Dame 30 seconds in).
Facial recognition technologies are nothing new in themselves and have been in use in airports for years, but getting them right in visual search is and remains an elusive, yet valuable prize.