Spatial Queries on Compressed Raster Images : How to get the Best of Both Worlds

The maintenance of large raster images, such as satellite pictures in geographic information systems, is still a major performance bottleneck. For reasons of storage space, images in a collection are maintained in compressed form. An operation on an image can be performed by rst decompressing the compressed version. This, however, can be a major source of ineeciency, because the entire image needs to be accessed even though a small part of it might suuce to answer the query. We propose to perform spatial queries directly on the compressed version of the image. With current compression algorithms, this cannot be done eeciently. We therefore propose a new compression technique that allows for the subsequent use of a spatial data structure to guide a search. In response to a range query, our algorithm delivers a compressed partial image; this response may be sent over a network as it is, or it may be decompressed for further processing. We have implemented the new algorithm, and we have compared it with standard and sophisticated compression algorithms (that do not support spatial operations). Our performance evaluation with satellite images shows that the new algorithm not only supports spatial queries, but is also competitive in terms of the compression that it achieves.