In this paper, we describe an image database system that can retrieve images that are similar to a query image. The query image is processed to extract information that is matched against an index to provide pointers to similar images. The salient feature of the system is that the index is developed from the JPEG-compressed images without first having to uncompress them. We use a similarity measure to determine the difference between the JPEG coefficients of the query image to those of the images in the database index. The index itself is developed by partitioning the JPEG-compressed image into a quad-tree structure and computing certain characteristics for each quad at each level in the tree. Each quad representation is stored in tables corresponding to the quad size in a relational database that can be queried by conventional means. Given a query image, or image segment, the corresponding quad tree-based representation is computed. Then, each of the quad representations in the query is used to search the index for the matching quads in the database. Initial experiments with the index have provided encouraging results. The system outputs a set of ranked images in the database with respect to the query using the similarity measure, and can be limited to output a specified number by changing the threshold match.
Image Databases, Indexing, Retrieval, JPEG compression