When can a photo be trusted?

Fourandsix on TwitterFollow @Fourand6
« Right Lane MUST Right Left | Main | How Precise is Your Forensic Analysis? (part 2) »

No Location Metadata? Ask Google Street View.

In an earlier post (“No Location Metadata? No Problem.”) I described a technique that can narrow the likely location where an image was taken using only the visual image content. This technique leveraged large geo-tagged photo collections to find images that match in overall appearance. Here I will describe a related technique that leverages a potentialy richer source of images from Google Street View images.

Photo sharing sites such as Flickr contain an enormous amount of geo-tagged images from around the world that can be used for image-based geo-location. These image collections, however, tend to be biased towards images of famous landmarks (the Statue of Liberty, the Eiffel Tower, the Great Pyramids, etc.). As a result, these photo collections may be less likely to provide matches against more mundane images. Repositories such as Google Street View, however, consist of an enormous number of street-level images that have a greater likelihood to capture location specific features of a city.

The goal of image-based geo-location is to extract visual elements that are frequently occuring within a given location and distinct to that location (for example cars are frequently occuring in Paris but not distinct to Paris, while the Eiffel Tower is highly distinct to Paris, but not frequently occuring.) Street View images have greater potential to yield such a desired set of visual features. In addition, Street View images are, by their very nature, guaranteed to be geo-tagged. A recent paper describes how to automatically extract such features and to use these features for geo-location (Carl Doersch, Saurabh Singh, Abhinav Gupta, Josef Sivic, and Alexei A. Efros. What Makes Paris Look like Paris? ACM Transactions on Graphics 31(3): 2012.)

Shown below, for example, are the discovered distinct features for Paris, London, San Francisco, and Boston extracted from a large collection of Street View images. Note that these correspond to fairly mundane features such as windows, street lamps, and other distinct architectural features that are likely to appear in many images.

Purely visual-based geo-location holds the potential to be incredibly useful to a forensic analyst. Photo-sharing websites such as Flickr and street-level photo repositories such as Google Street View contain an enormous amount of information that can be exploited. Street View images, in particular, are more likely to contain common and distinct features that can be compared against in order to determine the location where an image was taken.

[Credit: C. Doersch, S. Singh, A. Gupta, J. Sivic, and A. Efros. “What Makes Paris Look like Paris?” ACM Transactions on Graphics, 31(3): 2012.]

PrintView Printer Friendly Version

Reader Comments

There are no comments for this journal entry. To create a new comment, use the form below.

PostPost a New Comment

Enter your information below to add a new comment.
Author Email (optional):
Author URL (optional):
Some HTML allowed: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <code> <em> <i> <strike> <strong>