Daniel Arribas-Bel at Catapult: ‘When trying to understand something as complex as the city, every bit of data helps create a better picture. Researchers, practitioners and policymakers gather as much information as they can to represent every aspect of their city – from noise levels captured by open-source sensors and the study of social isolation using tweets to where the latest hipster coffee shop has opened – exploration and creativity seem to have no limits.
But what about imagery?
You might well ask, what type of images? How do you analyse them? What’s the point anyway?
Let’s start with the why. Images contain visual cues that encode a host of socio-economic information. Imagine a picture of a street with potholes outside a derelict house next to a burnt out car. It may be easy to make some fairly sweeping assumptions about the average income of its resident population. Or the image of a street with a trendy barber-shop next door to a coffee-shop with bare concrete feature walls on one side, and an independent record shop on the other. Again, it may be possible to describe the character of this area.
These are just some of the many kinds of signals embedded in image data. In fact, there is entire literature in geography and sociology that document these associations (see, for example, Cityscapes by Daniel Aaron Silver and Terry Nichols Clark for a sociology approach and The Predictive Postcode by Richard Webber and Roger Burrows for a geography perspective). Imagine if we could figure out ways to condense such information into formal descriptors of cities that help us measure aspects that traditional datasets can’t, or to update them more frequently than standard sources currently allow…(More)”.