We’ve already seen image search engines that claim to be able to find a particular image on the web, and services such as PicScout’s Image Tracker seem to have had considerable success in tracking usage for the big agencies who can afford to use it. Because it relies on a “fingerprint” created from the image rather than a watermark, it claims to be able to spot even images that have been cropped or manipulated.
TinEye is another “reverse image search engine” with the big advantage of currently being free to use (although they intend to add some extra paid-for services at some time. Tineye is very easy to use – just right click on any web image and select “Search Image on TinEye.” The only problem is that when I’ve tried it with my own and other peoples images that I know are on several places in the web it often hasn’t found them, though it is a lot better now than when I first tried it.
There are some examples of what it can do on the site. At the moment they claim to have scanned 1,143,177,077 images from the web; I don’t know how many there are in all, but the largest figure I’ve found on a single search on Google Images is 1,600,000,000.
Idée also have a PixID service which can identify usage of images in both print and on-line for “editorial, celebrity and entertainment firms, news wire services” and others, and is I imagine relatively expensive, as well as Pixmiliar which looks for similar images.
Similar to this is Gazopa from Hitachi America which is now in a public beta and you can use it on the web or download it for your iPhone should you have one. It lets you choose an image on your computer on on any web site, upload it and it will then find similar images. Possibly.
As I’m sitting here feeling sorry for myself just starting to recover (I hope) from swine flu – the letter offering me the vaccine came this morning, very much too late – I had time to try it.
The first few pictures – typical things I take like demonstrations – it could obviously not get it’s head round at all, so I thought hard and tried to give it something easy. How about one of London’s most recognisable buildings, Tower Bridge? And a picture that showed its very distinctive shape clearly:
and was rewarded with my first (and only) success. Of the 30 images on the first page of the thousand it selected there were actually 4 of Tower Bridge. I tried with some other well-known buildings from London, and other cities with no luck. A night image of the National Theatre did return night images, but not of anything remotely similar.
The bell-tower and cathedral in Brasilia doesn’t look to me much like the Japanese woman or the polar bear that GazoPa produced in response and although there was one other church it was in a very different style. The still ife image of two bottles was perhaps a little more understandable.
GazoPa also has a face recognition feature, so I thought I’d try that and fed it a picture of Tony Benn, complete with pipe, thinking there must be many thousands of similar images on the web (including hundreds of mine.) Searching normally perhaps the most interesting match – at third closest was of what looked like an amateur burlesque dancer, though further down the list were three men with pipes but otherwise no resemblance.
Switching to ‘Face’ mode gave a different set of matches, with several of the same kissing wedding couple, several women between five and ninety and a small child in a push-along cart. But perhaps those carved heads of four US Presidents on Mount Rushmore were a nod to his political nature. To be fair there was just one of the first 30 pictures that did bear a very slight resemblance to the man, and who was looking in a similar direction, but I think the matching algorithms still need a little improvement.
Just to give it a fair chance I thought I’d also try it on Tariq Ali who had been standing next to him on the plinth in Trafalgar Square, a couple of weeks ago. Perhaps I can see a very vague resemblance to a cat, but the pig just isn’t fair! And ‘Face’ mode doesn’t make anything better.
I’ve not posted any of the results here – as if PicScout picks up any from its clients it could bankrupt me. But you are welcome to try any of the images here – or on My London Diary – on Gazopa. If you find any really good examples perhaps you could post them as a comment – just the URL of my image and any comment you want to make would be fine.
