The utility of the co-occurrence matrix to extract slum areas from VHR imagery

Many cities in developing countries lack detailed information on the emergence and growth of highly dynamic slum developments. Available statistical data are often aggregated to large administrative units that are heterogeneous and geographically rather meaningless in terms of pro-poor policy development. Such general base information neither allows a spatially disaggregated analysis of deprivations nor are settlement dynamics easily monitored, while slums are rapidly developing in particular in megacities. This paper explores the utility of the co-occurrence matrix (GLCM) and NDVI to distinguish between slums and formal built-up areas in very high spatial and spectral resolution satellite imagery (i.e., 8-Band images of WorldView-2). For this study, an East-West cross-section of Mumbai in India was used. We employed image segmentation to extract homogenous urban patches (HUPs) for which the information extracted from the GLCM was aggregated. The result was evaluated using collected ground-truth information and visual image interpretation. The results showed that the variance of the GLCM combined with the NDVI separate formal built-up and slum areas very well (overall accuracy of 86.7%).