Researchers Develop Machine Learning System That Identifies Cities By Architecture

The test of Artificial Intelligence lies in emulating the human brain perfectly, but doing a better job at it. Researchers at Carnegie Mellon University and INRIA/Ecole Normale Superieure have developed a machine learning system that can identify cities, by looking at their pictures. This U.S.-French visual data mining project exemplifies the emerging field of visual data mining, which in context is far more difficult to implement than text or number mining.

#-Link-Snipped-#

As a part of the research, the machine learning system was familiarized with 40,000 Google Street View images of Paris, London, New York, and Barcelona, as well as eight other cities to pick the distinctive elements from each city. Features such as street signs, balconies, and lampposts of Paris made up for unique elements to identify the city through images. Though when it comes to U.S. cities, the machine learning system couldn't find precision due to the cross-cultural nature and lack of a single prevailing influence in U.S.

The machine learning system is capable of scanning through millions of image patches automatically as a part of data mining process. Researchers believe that this machine learning system can find applications in visualizing the geo-informative features for the entire planet, in the long term. Meanwhile, such a system can be used by film look designers to emulate the feel of a city.

https://youtu.be/s5-30NKSwo8

Source: #-Link-Snipped-# Image Credit: #-Link-Snipped-#