| Abstract | Forthcoming high resolution satellite sensors will generate monochrome images when operating in their highest spatial resolution modes. As a result, conventional image interpretation algorithms, such as
maximum likelihood classification, will be of limited applicability because of the paucity of spectral dimensionality. Instead, new feature and object information extraction methods, grounded primarily on spatial and contextual reasoning will be
required. To support these requirements, a testbed system has been developed to assess the utility of various spatial/spectral/contextual attribute combinations in the recognition of common features such as roads and buildings. A goal-driven, rule
based approach is proposed to analyze segmented renditions of imagery with the goals of delineating such objects and capturing their cartographic characteristics. Details of the segmentation, attribute processing, ground truth processing and rules
formulation are described. Example recognition modules to support residential street recognition are developed and assessed using 2-metre resolution aerial photography. |