An interesting aspect of spatial image analysis and OBIA in remote sensing is the theoretical consequences, which require further study. I suggest a theoretical framework that is a response to Bibby and Shepherd’s (2000) call for a new ontology in the application of GIS and remote sensing to land use issues. In their examination of the ontology of remote sensing and GIS, Bibby and Shepherd (2000) claimed that the philosopher Julius Moravcsik’s (1975) fourfold view of the meaning of objects has been conflated to only the Formal dimension in GIS, or ‘‘...physical attributes that distinguish one kind from another’’ (Bibby & Shepherd, 2000 p. 584). The other three dimensions are the Constitutive (the composition of things), the Telic (the function and purpose), and the Agentive (the processes that produce things). These dimensions are of interest to a wide range of users of remote sensing data, particularly those within the social sciences and UPM.
Spatial image analysis in the form of OBIA can incorporate the Constitutive dimension into the already inherent Formal dimension (physical attributes that discern one thing from another, i.e., electro-magnetic reflection), and the combination reflects aspects of the Telic dimension, which is of interest to the strategic scale of UPM. The main theoretical motivation behind using WICS in urban analysis is the principle that the local configuration (the Constitutive dimension) of materials with different spectral signatures, such as asphalt, grass or concrete (the Formal dimension), can be considered a visual fingerprint of function, planning and use (the Telic dimension). Extracting these fingerprints requires going beyond the scale of the individual building and field of grass and aggregating them into generalised concepts, such as classes of
the built environment. The WICS procedure comprises four steps (Fig. 1). Step 1 reduces the input data, for example, a multispectral satellite image, to a manageable size. This reduction could be performed through supervised or unsupervised spectral classification. Data reduction is necessary because the next step constitutes calculating distances between different pixels based on the spectral classes to which they belong. Step 2 consists of calculating the nearest-neighbour distance to all of the spectral classes for each individual pixel in the image. These distances are then used to create a contextual feature vector in the next step. A distance threshold can be applied to limit the effect of extreme values.
Step 3 creates a contextual feature vector for each pixel in the image. A contextual feature vector consists of the geographical nearest-neighbour distance from a specific pixel to pixels of all