Google is developing technology that can pinpoint the location of any photo using nothing but the image's pixels.
Many services can pinpoint where a photograph was taken using the hidden data which is often attached to most digital photographs.
However Google wants to go a step further and has developed a system that can determine where an image was taken by scanning the landscape.
Photos are broken down to the pixel-level, and Google's systems then try to cross-reference these with its gigantic image library to see if it can match it.
The sophisticated artificial intelligence program call PlaNet has been fed 90 million geotagged images from around the world.
They divided the world into a grid of 26,000 different sized sectors depending on how many images tend to be taken in different locations.
In a trial using 2.3 million images, PlaNet managed to determine the country of origin 28.4% of the time, and the correct continent in 48% of cases.
The test numbers are significantly better than humans performing the same test.
Google believes that with continued testing, PlaNet has the potential to get incrementally more intelligent.
Project leader Tobias Weyand said: "We think PlaNet has an advantage over humans because it has seen many more places than any human can ever visit and has learned subtle cues of different scenes that are even hard for a well-travelled human to distinguish."
0 nhận xét:
Đăng nhận xét