> It seems that the thematic folders and the file names may be ok
> descriptive tag sources to start with.  Perhaps you could try to identify
> patterns to extract information for tags (i.e., "hall", "committee",
> "holiday", etc.) You could traverse the file system, and use the Google
> Data API for Picasa, to do an initial upload with tags generated from those
> folders and file name patterns.  You will need *some* kind of user input to
> get to more detail.
> So, are you asking whether it's possible for software to generate metadata
> for what is pictured in a photograph without user input?

Not trying to avoid user input at all -- just trying to make it efficient.
The number of photos is such that hand keying everything into GUI forms or
even spreadsheets is a nonstarter. The idea here is to traverse the file
system, automatically add metadata based on derived info such as path to
existing metadata found in the image (if it is there), and then supply an
interface that allows people to assign/edit metadata easily. The reason I'm
interested in facial recognition is that automated assistance identifying
significant people is of high value to the institution and will help
prioritize work.

> I think you're on the right track by using Picasa.  I think it may be
> easier for people in those departmental units to help tag people, and add
> descriptions through that interface.  Then you can write some kind of
> script that slurps them into your repository.

That's the basic idea. Unfortunately, I can't upload data for processing
due to legal (loads of confidential and HIPAA protected info) and technical
reasons (tens of thousands of images to process, practically all I've seen
so far are over 100MB). But it sounds like Picasa could be part of a