Hello!
This month’s newsletter is a little different from our usual format. We wanted to tell you about something exciting that we’ve been working on with Manchester Libraries.
I Still Haven’t Found What I’m Looking For
Searching a collection of images, and particularly photographs, can be a frustrating experience. What you’re looking for might not be there at all – or it might be there, but simply not catalogued, described or tagged with the words you’re using to find it.
Manual subject indexing helps but it’s a long way from a perfect method of matching searches to results.
We decided to even the odds a little in researchers’ favour, using this one weird trick (actually a very simple AI-enabled process).
Automatic for the people
Mark 1 Computer, Manchester University
In 2024, we launched Manchester Libraries new website with over 100,000 photos from Manchester’s history. It’s an incredibly rich resource, but most image records contained only a title or a brief description. That meant poor alt text for screen readers, and also poor search results that didn’t match image titles.
So we brought in a little automated help. Manchester Libraries share a CIIM (our central data hub) with several other Manchester institutions. We used this CIIM to send each image to Microsoft’s Azure Vision. Azure ran three tasks on each image:
- Description: creating a mid-length caption
- Tagging: creating subject tags for each photo
- Optical character recognition: spotting any text in the image and indexing it
and then sent them whizzing right back to the CIIM with the new data attached to the object record.
Needle in a haystack
Only some of the outputs from this process, like the subject tags, appear directly on a record page. So when you find this picture of a baby you can quickly find other pictures of babies.
But when you search, the power of the underlying OCR’d text kicks in. It’s useful for streetscapes, where manually cataloguing every shop or building isn’t practical. Look for bookshops, or search for Manchester & District Housing Association Limited, and you’ll find housing with their billboards outside.
Suck it and see
Why not dive into the Manchester collections and see what you can find?
- Remember an old advertising slogan
- Locate a historical chippy
- Stock up on Vimto
- Find Guy Fawkes in both text and image
- Dixons? Squint (or zoom) and you’ll see it
- Search for a name on a war memorial
Water no get enemy
And at a time of all-round confusion about ‘AI’ and ‘generative search’ it’s important for us to say that this was a one-time controlled process (images in, and data out!) so your improved searches don’t use any more water or energy, and it was a fixed cost for us and Manchester Libraries.
Reach out (we’ll be there)
Tell us if the search helped you find something remarkable that you wouldn’t otherwise have stumbled across.
And if you’re interested in doing something like this with your own collections, then get in touch!
Until then,
Mike, Jeremy and the TMP team.