National Aeronautics and Space Administration

Identifying occurrences of objects of interest in a remotely sensed digital image and finding similar objects in a database of comparable imagery usually involves a high-level semantic description based on visual interpretation of every image in the database. This work proposes a similarity search approach wherein the user identifies an object of interest, spatial and spectral characteristics of the object are calculated, and the result is compared to a database of these same calculations that have been performed on all images in the database. The spatial extent of the object of interest is approximated using a region quadtree decomposition of the image. Spatial indices such as fractal dimension, lacunarity, and Moran’s I index of spatial autocorrelation, along with spectral characteristics expressed as histograms of each band’s gray scale values are matched against a set of these same indices that have been previously calculated for all images in a database. The sum of squared differences between the indices calculated for the quads that form theobject of interest and the same quads in the database yields a ranked list of images that have similar characteristics. The retrieval success rate is highly dependent on the configuration of quads used to define the object of interest and the nature of the object itself. Objects such as a lake shoreline are best retrieved using the gray scale histogram, while urban features that are characterized by their texture are more accurately retrieved using indices such as fractal dimension or lacunarity. 1.