Detecting Urban Social Inequality from Space using Computer Vision (and Chlorophyll)
In his recent blog post, Tim De Chant notes that poor neighborhoods can often be distinguished from richer ones by observing the amount of tree cover in satellite photography.
I wonder: can such a perceived visual-social correlation be confirmed algorithmically? And if so, to what extent does urban forest predict socioeconomic factors?
This software proof-of-concept (built with Processing) partitions satellite images into portions of "greenscape" and "hardscape." Thus, a given geographic region can be assigned a "percent greenscape", as indicated by the bar graph in the center.
Supposing that these green pixels correspond to trees, grass, and shrubbery, their proportion may serve as a useful proxy for population density and other quality of life indicators. By comparing this value with known demographic data for various regions, I will determine whether this rudimentary computer vision approach serves as an adequate measure of human demographic patterns.
(This early software already highlights one challenge of the method: non-normalized source imagery of variable quality, as seen in the Chaoyang example where much of the tree cover goes undetected.)
I've discovered a small literature (Jensen et al 2004) already published on urban forest and demography correlation using remote sensing, and with promising results. It will be interesting to see whether this can be further generalized.
June 4, 2012