InStreet 2 is an algorithm that allows to pinpoint a location where a photo was taken just based on the content of the photograph. Algorithms analyses the visual similarity between the query photo and the reference database of examples.
Imagine a situation when you have a photo taken in an urban area, but you do not have any additional metadata – no description nor geographic coordinates but you need to know where the photo was taken. Maybe the photograph is an evidence in a case? Maybe you found the photo in the Internet and are just curious? Or maybe you own a social network and wish to enable new innovative services?
When using the InStreet 2 algorithm the user has to provide the photo as a query and select a distinct architectural detail (e.g. a specific window layout or a unique shape of the entrance) or a text on a building (e. g. a banner). Algorithm compares the selected detail with the reference database and returns a set of best hits together with their geographical location. The user has to browse through the list and select a match.
InStreet 2 is an evolution of the award winning INSTREET (Gold Medal at the INNOVA fair in Brussels, medal of the French Inventors’ Union, named one of the “Inventions of the Year” by a major newspaper “Gazeta Wyborcza). The new, enhanced version of the algorithm utilizes not only visual similarity of the images but also analyses texts visible in the photographs.
Link to Instreet 2 demo: