Overview

The ability to effectively query document image databases is a problem which presents interesting challenges primarily because of the rich structure of the underlying documents. Unlike traditional databases in which structures and relationships between elements are well defined, an unprocessed document image contains a wealth of information, but is only represented initially by a set of pixels. For traditional databases, the underlying structure allows the user to form queries on specific fields or relationships between them, and retrieve a list of entries which match a specific query. Ideally, image databases should allow the same basic level of access, although it is often difficult to define queries in a quantitative way to facilitate such operations.

Work has recently begun on a joint project between the Universities of Maryland and Oulu on the development of a system for Intelligent Document Image Retrieval (IDIR). The IDIR architecture will provide close connections with, and utilization of, document analysis and image processing techniques, advanced computing and networking, and current approaches to database management. The system design consists of aggressively modularized components to enhance the development of individual parts which are used in the complete solution and include: Interface specifications, multipurpose feature extraction, an integrated efficient query language, physical retrieval from an object-oriented database, and delivery of retrieved objects.

The main technical accomplishment of this project is that we have developed methods for spatial indexing of document components and defined measures of document similarity based on these methods. This essentially allows us to define document image similarity based on structure. We have used the techniques query small databases for image that appear'' similar to a give query image with respect to structure. We can demonstrate its effectiveness for identifying documents such as title pages, bibliographics and advertisements by example.

Project Status

Recent work has focused on the development of spatial indexing techniques for document images. The current plan is to incorporate these indexing techniques and test the entire system on heterogeneous databases. Furthermore, we are integrating other indexing mechanisms such as imperfect OCR and the results of page classification.