How to determe the geographic location of a photograph
February 18, 2005 8:56 AM Subscribe
How can one (easily?) identify a location based upon a single image?
I have on my desk a nice tranquil image of a beach/cove - how can I go about determining where the place is?
There's been a very recent topic here with a similar request to identify an aerial photo, but just asking people if they recognise somewhere seems inherently... um... inelegant.
Is that the only way? Are there image resources that are categorised (and presumably meta-data'd) so that one can readily search for certain criteria within an image? Would that even be possible? It sounds like a huge undertaking...
Is my only solution to scan the image, and post it to as wide an audience as I can find? (Or perhaps a smaller, more select audience of knowledgeable people like on here ;-) )
I've put this in the "computers & internet" category, as I'm (optimistically) convinced that there must be a solution that is facilitated by computing!
I have on my desk a nice tranquil image of a beach/cove - how can I go about determining where the place is?
There's been a very recent topic here with a similar request to identify an aerial photo, but just asking people if they recognise somewhere seems inherently... um... inelegant.
Is that the only way? Are there image resources that are categorised (and presumably meta-data'd) so that one can readily search for certain criteria within an image? Would that even be possible? It sounds like a huge undertaking...
Is my only solution to scan the image, and post it to as wide an audience as I can find? (Or perhaps a smaller, more select audience of knowledgeable people like on here ;-) )
I've put this in the "computers & internet" category, as I'm (optimistically) convinced that there must be a solution that is facilitated by computing!
Asking for others to identify your location neatly chops the problem into small pieces. (Don't you consider net communications 'computing'?) Trying to identify on your own your particular beach/cove from photo archives could be quite a long, burdensome task.
Vid: on his desk, not his desktop
posted by mischief at 10:01 AM on February 18, 2005
Vid: on his desk, not his desktop
posted by mischief at 10:01 AM on February 18, 2005
If it's a stock photo, you could also try Corbis.com. You can search on various keywords ("horizontal beach palm water" or something, for example).
posted by occhiblu at 10:46 AM on February 18, 2005
posted by occhiblu at 10:46 AM on February 18, 2005
I don't think there's an automatic way if that's what you're getting at. Pictures are hard to describe in a manner meaningful to a computer. A lot of the details that people clue in on in order to identify a specific aerial view are abstract. Examples of this might be "in the upper left there appears to be some sort of arena" or "that looks like an airport".
I suppose you could produce something that could often do it based on artificial features. An edge detection/segmentation algorithm could look at the roads for instance and use it to throw out chunks of the database. Maybe identify all the roads in the picture and label them "Road 0 through Road N" and then use data mining to find all occurrences of a road intersecting another road at an angle of 15 degrees and so on.
You'd need a huge comprehensive database of artificial surface features though.
posted by substrate at 10:59 AM on February 18, 2005
I suppose you could produce something that could often do it based on artificial features. An edge detection/segmentation algorithm could look at the roads for instance and use it to throw out chunks of the database. Maybe identify all the roads in the picture and label them "Road 0 through Road N" and then use data mining to find all occurrences of a road intersecting another road at an angle of 15 degrees and so on.
You'd need a huge comprehensive database of artificial surface features though.
posted by substrate at 10:59 AM on February 18, 2005
occhiblu, that's why I love this place, all kinds of wonderful discoveries, thanks for the Corbis link!
posted by hardcode at 11:23 AM on February 18, 2005
posted by hardcode at 11:23 AM on February 18, 2005
This thread is closed to new comments.
posted by kokogiak at 9:06 AM on February 18, 2005