Casinos have funded a lot of facial recognition work, maybe they will pay for general object recognition. It’s a similar problem to computer assisted diagnosis and target detection, and those have a lot of funding (but not a whole lot results)

All the recognition software will likely be complex and will be run on the photos and tag them accordingly rather than run in real time, so there will still be keywords maybe just auto generated ones.

So I agree with you tim, given enough time and money we won’t need keywords, but one has to ask how long are you going to live?

I think there will be a lot of assistance with keywords.
Automated GPS tagging will help tag with locations (and may be aid object recognition)
Facial recognition will help with who is in the picture (though you will need to train it in the short term — here’s a pic of my dad, find other pics of my dad and tag them)
predominant colors can be done
And image similarity could be done so that it could auto suggest keywords from similar photos in your collection.

Google has an effort to get users to tag photos in their image search, and they are trying to develop algorithms that would apply those tags to other similar images, and thus be searchable, so maybe there is a way of leveraging all the images that are on web pages with descriptive captions, query the db for a similar image (not trying to figure out what it is, just similar) and then apply those “potential keywords” to images so that when you search for images you get exact matches and then maybe matches.

One analog that makes me think this all might be possible is that a lot of people don’t use bookmarks anymore, they just type their query in to google.

Also along this vein, you have people that ask for the image “with the thing running down the middle” which is hard for humans unless they were recently talking about that image. Though john nack of adobe points out a tool that lets you look up chinese characters by drawing them http://blogs.adobe.com/jnack/2007/12/trajan_the_hack.html so maybe you could draw salient portions of the image and find it.

Bryn cited the examples of “tropical plants” and “flying goose”, both of which I think, given enough time, could be figured out by image recognition somewhere down the line. And facial recognition of emotions is something that has been done with modest success.

In any event, I think it is a very intriguing topic, and suspect it is something that will get increasing attention as time goes on (lots of time, admittedly).

One is noting species names. I doubt that recognition technology can distinguish between shorebird species, or warbler species, or different skipper (butterfly) species. People have a hard time – I can only imagine the expert system required.

Another is special file attributes. If you keyword a file as “portfolio” you can return a list of those files. You can include this in a file name, but that can complicate backups if file names change.

Another case of special attributes is the file version you prefer to print, if you have multiple similarly named files. You can store files like this in one location – that’s a possibility with portfolio images as well – but you might want more flexibility than that.

And then there’s the question of whether operating system fetch/query technologies will satisfy all the use cases of digitial asset management – I wonder about that.

Further more, I’m not sure we’ll ever have image recognition software that can tell you if a picture is “melancholy” or is an appropriate picture for metaphors or phrases (picture of businessman standing in line associated with “early bird gets the worm?”)