Data modeling and feature extraction for image databases

Current image retrieval systems have many important limitations. Many are specialized for a particular domain of images, and are not applicable to other image domains. The more general systems treat all images uniformly. Consequently, the power of their query facility is limited to color, texture, shape, and other features that are common to all images, with no deeper understanding of the structure of a given image. Few systems (if any) have addressed the issue of scalability with respect to the size of the image collection and with respect to the underlying techniques. There are two communities that can contribute to image databases: Computer Vision(7) and Database Systems. In this paper we focus on the database side of the issue. We consider how to design a database system that supports a rich class of content-based queries on image collections, scales with collection size, and can easily incorporate future advances in computer vision. This paper outlines one approach, in the form of the design, implementation and testing of an image database system called PIQ. The main contributions we make are: 1. The idea of a data model for describing image data. This is coupled with an Object Model Description Language (OMDL) for describing image domains. We introduce our first cut at such a data model and its description language. 2. The ''Feature Extraction Manager''. This is a general algorithm for extracting features from images that utilizes the data model and any computer vision algorithms given to it and manages the search for objects in each of the given images. 3. A demonstration of the power of a query language that is built on top of an image data model. 4. A system design with an extensibility feature that can incorporate new feature extraction algorithms. To our knowledge, this is the first 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 OMDL on various aspects of the system, and experimentally demonstrate some major benefits of this approach. In particular, we show how very large image sets can be effectively queried-using meaningful, domain-specific 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.