Joint spatial-spectral indexing for image retrieval

Processing images directly in the compressed domain is very beneficial from the standpoint of lowering the computational complexity. In this paper we present a method of extracting features directly from the DCT coefficients of JPEG compressed imagery for the purpose of aiding image retrieval from a database. Our approach is characterized by a quadtree-based spatial structuring of the image followed by calculation of DCT coefficient features at the various nodes. Features based on DC and energy of AC coefficients are shown to be effective for queries concerning color, edge and texture properties.