Exploring the discriminating power of texture in urban image analysis

Abstract

This paper presents some preliminary results from a series of investigations into the use of texture analysis in urban image understanding. High spatial resolution satellite imagery of urban areas contains much information that is not adequately exploited using per-pixel classification techniques. The principal hypothesis addressed is that detailed spatial features may be recognised by the analysis of urban morphological texture. Results from two analyses are reported. First, co-occurrence matrix measures of homogeneity are used on a Spot Panchromatic scene of Harare, Zimbabwe, to predict housing densities stored in a co-registered database. Second a Fourier domain statistic is developed to measure residential block density and is tested on a Spot panchromatic scene of Cardiff, Wales. The statistic is used to predict urban population counts stored in a co-registered population surface. The results demonstrate that useful morphological information can be extracted from Spot panchromatic images using such methods.