Morrigan · Video walkthrough · 2 min

Image geolocation, outdoors and indoors

Morrigan suggests where a photo was taken from what is in it: the landscape, the room, the fittings, how people are dressed. It ranks likely countries and regions; an investigator decides what to do with the ranking.

Recorded in the Morrigan web application, on the narrator's own photos. Two examples are not an accuracy figure: we measure that on your own material in a pilot. Captions in eleven languages, from the CC menu.

  1. A desert selfie with no landmarks. Morrigan names the United Arab Emirates and puts Sharjah, where it was taken, at the top of its list. Dubai and Abu Dhabi stay on the list: the same desert runs across those borders.

  2. A room with a wall socket, a clue that points to the UK but fits South Africa and Namibia too. Morrigan ranks Ireland first, which is right. The county at the top of its list is wrong, though it borders the right one.

  3. A model trained on a very large number of images, reading the whole frame: how a room is put together, its fittings, how people are dressed. It runs on-premise or air-gapped, so a pilot can use operational material.

Read the transcript

Part 1 · Outdoors

If you sometimes need to figure out where an image or a video was taken, but you don't have EXIF data, then you might want to use our geolocation tool. I'm going to walk you through a couple of examples, one outdoor and one indoor, and show you how it works. So today's first example is me with a couple of friends from the FBI and Home Office, somewhere in the world. Work trip. No landmarks, no convenient water bottles in the foreground. Hard to figure out where we are. But our AI model can figure out just from the image itself, just from the appearance of the image, where it might be produced. In this case, it's named not only the country, the UAE, but also the specific province. It is indeed taken in Sharjah province, although it is reasonably consistent with other provinces within the UAE, like Abu Dhabi or Dubai. Now you might be thinking, right, that's great.

Part 2 · Indoors

Works outdoors, works in the desert. What about indoors? So much crime happens indoors. What if I don't have an outside view? Alright, let's have a look at this image. It's indoors, and we can see a wall socket. Looks a bit like it might be from the UK. But then South Africa and Namibia also use the same socket types. Where in the world do you think it's from? Actually, this image was taken in a house in Ireland. The AI has correctly got Ireland as the most plausible country. Within Ireland, it's got the wrong county at the top of the list, but it's right next to the correct county.

Part 3 · How it works

How does it work? It's not magic. It's just using local knowledge. We've trained our model on a very large number of images, and it knows how homes and offices look around the world. It has pretty good knowledge of outdoor areas too. And it's looking at really everything in the image. You know, how the room is put together, how people are dressed, all that kind of thing. Best way to figure out if it works for you is to get a demo, or maybe even a free pilot. So you can use it on-premise, air-gapped. Use it on operational data. Thanks for watching. You'll see my contact details at the end of the video.

Try it on your own images

Image geolocation Try it on one of your own photos The public demo runs Morrigan on a single image and ranks the likely countries and regions. Free, no account. It narrows a search: it never gives an address, and not every image has enough in it. Open the geolocation demo → Media classification Grade an image and see where it was probably taken The classification demo proposes a severity for an image. For your first ten images a day, it also shows Morrigan's estimate of the country it was taken in. Open the classification demo →

Questions: [email protected]