Digital Image Access Retrieval. Edited by P. Bryan Heidorn and Beth Sandore. Urbana, 111.: Graduate School of Library and Information Science, University of Illinois at Urbana-Champaign, 1997. 191 p. ISBN 0–87845–100–5. (Clinic on Library Applications of Data Processing, ISSN 0069–4789).
Each of this book's eleven papers explores a different aspect of digital image access. The papers were originally presented at a 1996 symposium, the Thirty-Third Annual Clinic on Library Applications of Data Processing, 24-26 March 1996. Prepared by researchers and practitioners in various fields, they provide a forum for the exploration of creative solutions. Following the editors' lucid and informative introduction, the papers present digital image technology in its multi-faceted impact on libraries today and tomorrow.
The three papers in the first section (“Systems, Planning, and Implementation”) provide a framework for exploring the historical, technical, and management issues involved in digital image collection development. The six papers in the second section address the central issue of automatic and semi–automatic indexing of images. The subject of the final section's two papers is the use of digital images as replacements for deteriorating originals in the preservation process.
Image Databases: The First Decade, the Present, and the Future”, the keynote paper by Howard Besser, chronicles the relatively short history of digital imaging. The author shows how much has been accomplished within such a short time, but also points out the considerable work still left to be done. Besser was involved in library and museum image database development before many thought image databases were a practical consideration. He presents an eloquent synthesis of the technological innovation needed for images and of library service. Innovation in several critical technologies (including storage, telecommunications, and processing power) has fueled the current fast pace of development, and it is interwoven in his outline of the roles of library and information scientists. Besser identifies crucial areas for future work, as in standards and metadata development, image quality issues, and approaches to content-based retrieval.
Jennifer Trant explores the administrative, technical, and legal mechanisms for the eventual delivery of large high–quality collections of museum images to educational institutions in her paper, “Exploring New Models for Administering Intellectual Property: The Museum Educational Site Licensing (MESL) Project”. The MESL consortium is a two-year cooperative project of seven museums and seven universities. Its purpose is to establish standard practices to accommodate this new potential for the distribution of images. In describing it, Trant presents some models of rights management, namely, rights holders' collectives, brokerages, rights resellers, consortia, and locator services. She juxtaposes numerous and complex issues of legal and intellectual property rights associated with the migration of rights from original works to reproductions. The likelihood of theft is greatly increased because of the ease with which an exact copy of a digital image can be made and transported.
Lois Lunin's “The Big Picture: Selection and Design Issues for Image Information Systems” is the final paper of this section. It contains a valuable decision–making framework for anyone contemplating the development of an image database for any substantial collection. Examining many factors involved in the planning and design of an image information system, Lunin warns that it is easy to be preoccupied with the technology alone, and to neglect the broader planning required before a project begins, including an analysis of users' needs and the eventual desired functionality of the system. It is necessary to identify the users of the system, the images of interest and their sources, the property rights associated with them, short– and long–term hardware and software needs, standards, and costs.
The second section, “Automatic and Semi-Automatic Indexing”, is dedicated to the discussion of research and commercial systems that perform automatic content–based indexing of images. Each of its papers includes a theoretical discussion of the problem domain and the principles used to solve the problems in that domain. Making a unique contribution to progress in the field, each of the six projects addresses the central issue of automatic and semiautomatic indexing of images. In his “Content-Based Image Modeling and Retrieval”, Rajiv Mehrotra focuses on the thesis that an image database (or any type of database) is effective only if it models the real world attuned to the user's perceptions and needs, and if the representation of image content in a system is recognizable to users. Mehrotra tests his ideas in the prototype system, MUSEUM, which uses separate models for representing and processing two- and three–dimensional object shapes. Discussing the role of multimedia access in digital libraries, Ramesh Jain points out the difficulty of supporting queries at the semantic level when an integration of media is attempted because of the heterogeneity of formats. In his paper, “Visual Information Retrieval in Digital Libraries”, Jain introduces VIMSYS a data model that supports a hierarchical representation of images by using domain–independent physical characteristics of a scene, along with the domain–dependent features and relationships in the image.
“Efficient Techniques for Feature-Based Image/Video Access and Manipulation”, by Chang et al, centers on the parallel issues of image and video indexing, retrieval, and manipulation. The problem of heterogeneous unconstrained collections is addressed in VisualSEEK, a content-based image query system, and CVEPS (Compressed Video Editing and Parsing System). “Multi–media Analysis and Retrieval System (MARS) Project”, by Huang, Mehrotra, and Ramchandran, provides a rich discussion of current domain–independent automatic image indexing research. The richness and diversity of indexing techniques are exemplified by indexing based on global image color and texture, image segmentation, layout, shape descriptors, compression, and multimodal query integration. The authors present their own problem, i.e. how to efficiently integrate this broad spectrum of features into a single query. David Forsyth et al. address some of these issues, and others, in “Finding Pictures of Objects in Large Collections of Images”. The pre–coordination of feature sets to create object detectors, such as a horizon filter, is one of the most interesting features of the system. This paper presents three case studies: (1) the use of low–level color and texture properties for indexing and classification; (2) the use and exploitation of geometric constraints when the domain is limited; and (3) using domain–specific semantics to develop a filter for automatic identification. Rohini Srihari's “Using Speech Input for Image Interpretation, Annotation, and Retrieval” centers on the design and implementation of Show & Tell, by using the semantic output of a natural language processing system to direct image analysis and the object identification task. In this semi–automatic indexing and retrieval system, a human provides a verbal and gesture description of objects in the image, and this directs image processing. Both the natural language processing and the image processing are dependent on a domain–specific world model or ontology. Index terms for later retrieval are taken from annotations provided by humans and attached to the objects in the segmented images. Parallel techniques of using natural language to direct image processing are used in the Srihari group's MMVAR, a Multimodal System for Video Annotation and Retrieval which implements the parallel techniques of using natural language to direct image processing. Techniques used in Show & Tell and MMVAR demonstrate the advantages of integrating processing multiple modalities, and of exploiting rich domain–specific ontologies.
This reviewer agrees with the editors' comments: “Taken together, the papers in this section mark the forward edge of research in automatic content-based image indexing and retrieval. We can expect this to be an active area for research and commercialization for the next decade.” (p.9)
It is important to note that none of these systems solves the image database problem in its entirety. One of the significant research questions that remains to be explored is that of mapping from the low–level features are recognized by image processing techniques to the high–level and conceptual features of interest to humans. This question is not well addressed in any of the six papers, and the unspoken assumption seems to be that libraries do not have the resources to manually index the extensive digital image collections they are creating. It is noteworthy that two of the six federally–funded digital library research projects focus directly on image processing and retrieval systems. The contributions by Chang et al. and Forsyth et al.m represent research supported by the NSF/ARPA/NASA Digital Libraries Initiative Program.
The final section of this book contains two papers on preservation as it relates to digital media. In “Digital Imaging: Issues for Preservation and Access–, Meg Bellinger outlines the general issues for any preservation project, pointing out the fallacy of digital incorruptibility. Considering digitization projects as one aspect of a comprehensive preservation plan, Bellinger identifies these as some of the factors to consider: the quality of the digital image as a factor of resolution, authenticity, verification, bibliographical integrity, preserving and archiving digital media, and the obsolescence of equipment and standards. In “Preserving the Past: The Development of a Digital Historical Aerial Photography Archive”, Donald Luman discusses digitization and digital manipulation. He demonstrates the use of image processing to repair and enhance photography. How well digital representations of archival material can be merged with modern data is demonstrated by the merging of archival images with modern digital orthophotography to provide a three–dimensional view of the terrain of the past, which can then be compared with the current landscape.
Digital image access and retrieval represent the harnessing of multi–media information resources. This book brings together a balanced selection of articles to explore present issues in digital imaging access and retrieval. The timely theme of the book, the automatic and semi–automatic indexing models of images, is analogous to the natural language indexing of full text in the 1960s and 1970s. Understanding this central issue of the book is a necessary step in pushing us to the next stage of innovation, in order to provide conceptual indexing for humans. Consideration for the intellectual and logical indexing structure for image access and retrieval, and its eventual convergence with indexes of all types for retrieving and organizing any information, will indeed require new innovative exploration at the next level. The next level of indexing an individual image will be a daunting challenge, if a picture is worth a thousand words.
Gertrude S. Koh is Professor, Graduate School of Library and Information Science, Dominican University, River Forest, Illinois, U.S.A.
© 1998 Dominican University