Academic journal article Geographical Analysis

Urban Textural Analysis from Remote Sensor Data: Lacunarity Measurements Based on the Differential Box Counting Method

Academic journal article Geographical Analysis

Urban Textural Analysis from Remote Sensor Data: Lacunarity Measurements Based on the Differential Box Counting Method

Article excerpt

Lacunarity is related to the spatial distribution of gap or hole sizes. For low lacunarity, all gap sizes are the same and geometric objects are deemed homogeneous; conversely, for high lacunarity, gap sizes are variable and objects are therefore heterogeneous. Textures that are homogeneous at small scales can be quite heterogeneous at large scales and vice versa, and hence, lacunarity can be considered a scale-dependent measure of heterogeneity or texture. In this article, we use a lacunarity method based on a differential box counting approach to identify urban land-use and land-cover classes from satellite sensor data. Our methodology focuses on two different gliding box methods to compute lacunarity values and demonstrate a mirror extension approach for a local moving window. The extension approach overcomes, or at least minimizes, the boundary problem. The results from our study suggest that the overlapping box approach is more effective than the skipping box approach, but that there is no significant difference between window sizes. Our work represents a contribution to not only advances in textural and spatial metrics as used in remote-sensing pattern interpretation but also for broadening understanding of the computational geometry of nonlinear shape models of which lacunarity is the reciprocal of fractal theory.

Introduction

Despite the new generation of very high spatial resolution sensor data (IKONOS from 1999 and QuickBird from 2001), predicted improvements in classification accuracy of urban land covers (and subsequent inference of urban land use) have yet to materialize substantially (cf. Aplin 2003; Herold, Goldstein, and Clarke 2003). Much of the obstruction to quality information extraction is still due to the traditional limitations of classifying image data representing urban areas: the high spatial arrangement of complex urban features and how to configure multispectral responses from land cover features into organized urban land-use categories (Barr, Barnsley, and Steel 2004). When launched, the desired objective of high spatial resolution sensor data was for increased clarity of terrestrial features, especially urban objects, by reducing per-pixel spectral heterogeneity and thereby improving land cover identification. Clarity is certainly more evident in these finer-scale data than those from preceding sensors, but paradoxically this greater level of detail is also translated into many more unique per-pixel spectral combinations. For example, the residential land-use category can now be defined from much wider spectral variations, representing minute compositional mixtures of urban land covers, such as roads, houses, grasses, trees, bare soil, shrubs, and swimming pools, each conceivably a different residential land-use category. Following on, another limitation for improved information extraction from high spatial resolution sensor data is the reliance on techniques using traditional per-pixel spectral differentiation. To us this seems counterintuitive and we would like to see more neighborhood-related methods, using textural and spatial parameters when dealing with fine-resolution image data. Where traditional spectral approaches are designed to identify homogeneous features regardless of shape, textural and spatial algorithms measure both the variance within and the geometric configuration of whole urban objects, respectively (see Wu et al. 2000; Tullis and Jensen 2003; Herold, Goldstein, and Clarke 2005). As a contribution to the growing literature, we outline an object-based pattern recognition technique that accommodates the concept of lacunarity for characterizing the textural properties of urban land cover (and therefore inferring land use) from high spatial resolution image data. In doing so, we consolidate the utility of geometric models not only for image data but for all discrete and textural spatial representations (Zhao and Stough 2005). Indeed, the ability to characterize the shapes of individual and groups of objects is a rapid area of research in computational geometry and at the heart of the recent developments in object-based models in many geographic information system algorithms (Medda, Nijkamp, and Rietveld 1998; Wentz 2000). …

Search by... Author
Show... All Results Primary Sources Peer-reviewed

Oops!

An unknown error has occurred. Please click the button below to reload the page. If the problem persists, please try again in a little while.