Content-Based Queries in Image Databases

Current image retrieval systems have many important limitations. Many are specialized for a particular class of images and/or queries. The more general systems support relatively weak querying by content (e.g., by color, texture or shape, but with no deeper understanding of the structure of the image). Few (if any) have addressed the issue of truly large collections of images, and how the underlying techniques scale. There are many aspects to a DBMS supporting image retrieval by content. In this paper, we focus on a data model and give an example of a data deenition language (DDL) for image data, and demonstrate the gains to be had by incorporating such a DDL in a general-purpose image DBMS. Speciically, we make contributions in ve areas: (1) A proposal for a data model for images. (2) The use of DDL deenitions for guiding automatic feature extraction, using a constraint-based scheduling algorithm that calls upon a library of standard and specialized image analysis routines. (3) The use of extracted features (based on the data model and DDL deenitions) in representing and indexing large sets of images, and in query formulation and evaluation. (4) A system architecture that supports the use of specialized feature extraction algorithms, which may be independently developed in various important application domains, and may rely upon domain-speciic image analysis techniques. To our knowledge, this is the rst proposal for the use of a non-trivial data model (coupled with an image description language) for processing large sets of images in a DBMS. We discuss the impact of the data model and the DDL on various aspects of the system, and experimentally demonstrate some major beneets of this approach. In particular, we show how very large image sets can be eeectively queried | using meaningful, domain-speciic restrictions on the attributes and relationships of objects contained in images | with users providing input only on a per-collection, rather than a per-image, basis. We show that the approach is scalable, and demonstrate that content-based querying of very large collections of images using a domain-independent image DBMS is a viable goal.