A Hierarchical Object-oriented Approach for Extracting Residential Areas from High Resolution Imagery

Change detecting and timely updating of residential areas is widely recognized as one of the most challenging tasks for an operational GIS. High spatial resolution satellite data provide an efficient source of information. This paper describes a hierarchical object-oriented approach for extracting residential areas from IKONOS and SPOT5 images acquired in 2001 and 2004, respectively. Hierarchical object networks are formed through segmenting the image into non-overlapping and homogeneous regions (i.e. image objects) with separated fine and coarse resolutions, and constructing topologic relationship between them. The extraction of residential areas is implemented in three processing steps based on this object network. First, classify the image objects into several land cover classes such as roofs, green lands, roads, barren lands and water using image objects’ spectral, textural and spatial properties identified by the fine-resolution segmentation. The second step is performed based on the classification result of the first step for coarse-resolution segmentation. Only two classes are concerned here, namely, residential and non-residential. The decision to whether an image object belongs to residential is made by computing the percentage of ’roofs’ image objects derived from the first step. The third step merges all adjacent image objects that belong to residential class. All adjacent image objects that represent the same, or partially same, structure are merged into one new image object representing the whole residential area. The change of residential areas between the two image dates is then analyzed and discussed.