Semantic Video Analysis... or why I hate keywording

The Economist has an article about how computers are now beginning to be able to recognise the content of photographs and video. Modern point and shoot cameras now all have a 'face detection' mode to make sure they focus on the person you are trying to take a picture of and not the wall behind them. And some even now sport 'smile detection', only taking the picture once everyone is at full grin. Useful for granny's birthday party no doubt.

But the new technology being developed has a far more useful application. It may eventually allow the content of pictures and video to be determined automatically. Currently if you search for a picture online, e.g. at a stock agency, you can only find images if they have been carefully keyworded. These keywords are then stored as 'metadata' along side the actual image file.

As someone who submits images for stock I can vouch for how time consuming this can be, so I'm looking forward to the day when our computers can do it for us - even if it does mean I have to get my mouth around Semantic Video Analysis.