Remember when photography was about light and shade, composition and chemical magic? Now it’s mostly about batteries, megapixels and, increasingly, keywords.
The vexed question of how to easily and accurately keyword images so they may be quickly found in databases is one that increasingly exercises photographers. Solutions to the problem vary. Some agencies and individuals outsource keywording, often to developing countries, and we’ve even heard of a monastery in the USA that offers a commercial key-wording service: a sort of 21st century version of manuscript illumination.
Most, however, do the grunt work themselves, either staring at images until they go blind or literary inspiration strikes, or using programmes such as Image Info Toolkit to ease the pain. Some of course simply resort to keyword spamming, stealing keywords and text from other sources, a particularly popular practice amongst some Alamy contributors.
But in the auto-everything mentality of the modern world, the holy grail of keywording for many would be an automated keyword system. Step forward ALIPR: Automatic Linguistic Indexing of Pictures – Real Time, to use her full name. And yes, the developers really do claim their new baby is a girl.
ALIPR is the brainchild of Professors Jia Li and James Wang of Penn State University, and was officially launched on November 1. However in an attempt at success by association, the press release mentions that they started their work in 1995 at Stanford University, “where the Yahoo! and Google projects were started by their fellow schoolmates”. Well, quite: that’s a bit like announcing that you’re forming a new football club on the basis of having hung around back streets with Wayne Rooney a decade ago.
To be honest, ALIPR has a bit to do to catch up with Google and Yahoo. She only speaks English at the moment and has a rather limited vocabulary of a mere 332 words, although that’s admittedly more than many photographers. And she can only analyse colour images, which does rather leave a few well-known photographers out of the frame, not to mention the bulk of photographic history.
Some of ALIPR’s keyword suggestions are, to be kind, rather random. She correctly identifies George Bush – and a few distant relatives – as primates, but fails to return him as president: does she know something we don’t? And some of her conclusions are just plain wrong. Encouraged to search ALIPR extensively by Rob Galbraith’s promise of blush-making content [it’s a tough job but somebody’s got to do it] we were rewarded by a low angle image of a young woman almost wearing a thong. It’s the kind of image that lends itself to extensive, graphic and highly enjoyable keywording: the hours simply flew by. Yet strangely the official EPUK keywords were entirely different from ALIPR’s offering of “indoor, modern, man-made, interior, card, play, space _ shuttle, ocean, beach_flower, landscape, sky, plant”: perhaps she’s just shy.
Of course it’s unfair – but fun – to pick holes in ALIPR. As Li and Wang point out, ALIPR is in her infancy – still in her pram, so to speak – and their site offers visitors the opportunity to broaden baby’s vocabulary by making suggestions and uploading images. But that’s a development strategy that could lead to all sorts of problems, as Li and Wang may already be realising. “Please do not upload objectionable images,” they plead, seemingly forgetting that this is the internet, home of objectionable images.
After all ALIPR’s big ambition is to make as many images as possible easily findable on the internet. And to do that she’ll have to associate with the kind of images we only ever see when we check our bookmarks.
Primal Scream: objectionable images at ALIPR