Google Images - David Leonard

Google Images will try to identify the subject matter of an image you upload to it and offer you similar images

Monnow Bridge in Monmouth. Instantly recognised by Google Images (ignoring the text result suggesting it was in Lyon and the Getty Images result that thinks Monmouth is in England)

I have been scanning in photographic negatives taken by my father and came across one of an interesting-looking bridge. Wondering where it was, I offered it up to Google Images. The first (text) result that Google offered was wrong: I knew the bridge wasn’t in Lyon. But there then followed lots of “visually similar images”, from which it was easy to identify my image as being of the Monnow Bridge in Monmouth. I can’t help adding, though, that the Google result from Getty Images said that the bridge was in England (but that’s hardly Google’s fault).

Wondering how good Google Images is, I tried a few other examples:

It similarly had no problem in identifying Torquay Marina, but that was probably helped by the fact that my viewing angle and position when taking the image were probably something of a cliche.

Then I tried an image of Eduardo Paolozzi’s statue of Newton outside the British Library. Again, instant match. Clearly, Google Images could be really useful in identifying places and objects you’ve snapped on holiday and then, of course, going on to give you more information about it than you could possibly need.

Next, I thought I’d try a sterner test. Long ago I had been playing with an image of the Lincoln Imp taken from a postcard. I’d previously stripped off the area around the imp itself so that made things harder for Google by removing the context. The British part of me was definitely pleased that the American Google not only failed to identify the subject, but that most of its guesses involved Gollum from the films of Lord of The Rings. Hah, not such a know-it-all, then.

Then I decided to make it easier by doing a text search on “Lincoln Imp”, downloading an image of it, and then offering that image back for identification. This time it got it right. It was similarly instantly accurate when I offered it an image of myself that is on my website, but it failed when I offered a different image of myself. To a privacy freak like me, that is a profound relief! The conclusion at that stage was that it can instantly find an image that is already in its “memory banks” but that it isn’t quite as good at interpreting an image that doesn’t exactly match something that it already knows about.

O2 and Tall Ships — Google Images either didn’t spot the O2/Dome or chose to ignore it

With a hint of malicious glee, I then offered it a real challenge – a black and white version of a photo taken by me of the Millennium Dome (or “O2”, as it is known by the people who believe in building naming rights) with some tall ships featured more prominently than the Dome. Google’s results didn’t mention the Dome/O2 at all. It assumed the ships were the important feature. One up for the humans vs artificial intelligence, methinks.

Google Images is pretty impressive, but reassuringly imperfect. All you need to do to use it is:

Open the Google Search page
Click on the link labelled “Images” at the top right
Click on the icon of the camera that is now inside the search box
Either type in the full address of the image (its URL), click on “upload an image” to select one from your computer, or just drag an image into the box