Managing inhouse databases incorporating images

Michael Middleton

Wednesday, 25 August, 4.45pm

ABSTRACT:

INTRODUCTION

Well, of course we must come to terms with the contemporary application, particularly because of the influence that the multimedium is having on that lifeblood of information managers - database creation and development.

It was observed fairly recently (or in information technology terms, "some time ago"), that the success of multimedia information services depends upon four things(Bulick, 1990):

- Ubiquity of high-bandwidth networks;

- Adoption of standards for data representation compression,packaging and transport;

- User applications capable of handling different information formats;

- Development of a corpus of multimedia information and an

associated infrastructure for organising and searching it.

I will address each of these areas briefly. I will endeavour to pay most attention to classification and to retrieval, because whilst there is a considerable body of multimedia data and supporting technology now available, the central dilemma of the information intermediary - to organise, describe and produce databases in order to optimise retrieval - remains problematical for multimedia databases.

CREATION

In preparation for image storage, technical information must be created additional to the picture image itself. This may include information about the individual picture elements (horizontal and vertical pixel display), bits per pixel and colours assigned to bit combinations.

The image may be represented in a number of ways but we generally distinguish between a raster image which is a bitmapped matrix of the pixels, and a vector graphic image which is essentially the geometric definition of how the bitmap may be drawn.

Although vector graphic images must be turned into raster images for display, they are more compact to store, easier to edit and to "iconify".

Sound is essentially analog information, but as we know from compact disks it can effectively be represented digitally. This at least eases the considerable problem of synchronising sound with animation or moving images from databases.

COMPRESSION AND STORAGE

For example using the Australian video PAL standard of 25 frames per second, with each frame needing about 300 kilobytes without sound for good VGA resolution, means that full motion requires about 500 Megabytes per minute (Willie et al, 1993). Without compression, this would mean that a typical CDROM disk would hold about 1 to 2 minutes of video. Time varying compression standards such as MPEG, DVI and CD-I may obtain compression down to about 1% of original at recording and playback using the principal of recording differences within frames (intra-frame) and between adjacent frames (inter-frame) rather than storing the frames themselves.

Whilst CDROMs may therefore be used for archival and read-only databases and carry about an hour of compressed motion imagery, rewritable magnetic or magneto-optical mass storage is preferred for databases that are continuously maintained, as is often the case with in-house data. For example, the VTLS Infostation being developed in the library environment is based upon NeXT personal workstation hardware that incorporates 256 Megabyte magneto-optical disks (Lee, 1990).

TRANSPORT

Rapid development of publicly available databases, usually of a text orientation, has been contrasted with the lack of availability of numerical data despite the many and varied "in-house" collections of such data. Much of this could be ascribed to lack of network infrastructure and to the commercially sensitive nature of material. It is also attributable to a lack, until comparatively recently, of an appreciation by management that information can be used not just as a strategic resource for the internal benefit of the corporation, but also as an income producing resource for the corporation's financial benefit.

Now that network technology makes access to remote databases relatively straightforward, the distinction between in-house and external (or published) databases is likely to diminish, and be driven solely by management decisions about which parts of corporate data should not be made publicly available: how much of a view of the corporate databases may be made available for income-producing purposes without prejudicing company intelligence?

Allied to management considerations, is the increasing provision of distributed database architectures enabling a standard query approach to multiple databases which may therefore appear to a user to be one local database. The DOMINO system, for example permits linking of homogeneous databases stored on different devices such as WORM, CDROM and magnetic disks(Bordogna, 1990).

We can expect to see the process of standardised access to multiple remote systems supported further by implementation of the ANSI Z39.50 standard for interconnecting information retrieval systems. This will enable a user interface to be independent of a remote information retrieval system and to act as a client to multiple remote servers.

The ability to distribute multimedia information within organisations has been made possible by Local Area Networks. Bandwidth, the key to being able to move large amounts of data around is relatively cheap on LANs and makes in-house data access perfectly practicable.

Wide area transport of multimedia data is straightforward even on relatively narrow bandwidth lines providing that resolution requirements are not great, that compression and decompression processes are applied, and that the databases are transferred and then searched locally. However when real-time viewing of colour motion is required as in videoconferencing, the digital load goes into the Gigabit/second range, and requires Digital Signal Processors for effective compression/decompression on wide band optical fibre links.

INFORMATION RETRIEVAL

The object-oriented nature of databases that contain images has lent itself more to the associative information management system paradigm for searching. This approach which we know as hypertext or hypermedia differs from the established approaches in that the retrieval focus is divergent using links, rather than being convergent using inverted files or relations. It also provides a more open design philosophy, enabling database information structures to be content driven and based upon nodes or objects, rather than data elements or fields (Franklin, 1989).

The hypertext approach lends itself to graphic display of meta-information such as thesauri used for indexing databases. It presumes that the spatial metaphor will enable easier grasp of concept structures via node representation (Craven, 1992). It also has been put forward as an enhancement of text retrieval in the databases themselves.

However, whilst the hypertext approach to searching presents the technical solution to the information retrieval approach presaged by Vannevar Bush's 'memex' for fully linked data, it presents an expensive overhead in preparation time. For example, some preliminary studies (Al-Hawamdeh, 1991) have shown that there were disparate and inconclusive results obtained when comparing hypertext searching of fully linked textual documents with a more standard approach. However, the preparation for hypertext searching by building of the links for the full-text documents took of the order of 24 hours per document, much longer than preparation of documents for inverted file searching.

Nevertheless, the hypermedia approach to searching databases containing images has increasingly been used on the basis that non-linear retrieval may be applied with success to image searching. For example, in a numismatic database that contains text and graphics, where the graphics are both line drawings and scanned images, it has been found useful to identify a coin type then browse hypercard stacks (the Macintosh metaphor for hypermedia), to identify similar coins (Moline, 1991).

A principal concern with developers of hypertext systems has been the navigation problem, sometimes expressed as "getting lost in hyperspace". This refers to the potential for disorientation in a user pursuing a complex web of linked relationships in a document whose size cannot be approximated.

Attempts to deal with this have included constraining the extent to which links are provided, and enabling users to display to themselves in-context maps of labelled nodes with directed links. One experiment in this area (Gordon, 1992) has indicated that provision of maps had no significant effect on learning, where subjects were undergraduate students endeavouring to assimilate tutorial information.

It has been suggested (Bernstein, 1991) that there is no convincing evidence to suggest that disorientation is more significant in hypertext than in linear systems. With increasing familiarisation by users, better approaches to on-screen layout and typography by developers, and development of sophisticated navigational aids, we can expect to take advantage of hypertext to provide in-context help and surmount prejudices about large documents. Indeed users of a hypertext system by selecting their own links are able actively to shape the document that they read.

CLASSIFICATION

At present development is proceeding with databases that combine text and images in a linked manner. For example, to address the problem of retrieval of still-frame and motion picture documentation of the U.S. space flight program, NASA has developed a visual thesaurus (Seloff, 1990). This thesaurus began with about 200 images that are linked to text and also hierarchically linked to each other. The thesaurus is used for description of images obtained from the space flight program. It has been found that the visual approach is of most interest and use to inexperienced users and the text approach is preferred by experienced users.

Further work in the same area of space flight imagery has involved the development of a prototype system that integrates expert system and database technology (Ragusa, 1993).

It has been recognised that NASA classification specialists have developed rules of thumb for determining the relative importance of different objects in space shuttle pictures, and that retrieval specialists have developed an understanding of the logical sequence in which engineers seek required images. An expert system shell has been used to present classifiers with images linked by "hyperregions". If the classifiers then select a region, a database record (using dBASE III PLUS), is created automatically for thesaurus-based text description of the objects. The same image-based approach may be used for searching to enable information retrieval firstly from the linked database, then from the linked bit-mapped mass storage of the images themselves.

SUMMARY

In the area of information management, the principal endeavour that must now be applied is to the organisational and navigational structure and processes that provide the metadata support for the database itself.

REFERENCES

Bernstein, M. 1991 'Deeply intertwingled hypertext: the navigation problem reconsidered' Technical Communication 38(1) pp. 41-47.

Bordogna, G. et al 1990 'A system architecture for multimedia information retrieval', Journal of Information Science, 16, pp. 229-238.

Bulick, S. 1990 'Future prospects for network based multimedia information retrieval', Electronic Library 8(2) pp. 88-99.

Craven, T.C. 1992 'Concept relation structures and their graphic display', in: N.J. Williamson and M. Hudon, eds. Classification Research for Knowledge Representation and Organization, Elsevier, Amsterdam, pp. 49-59.

Franklin, C. 1989 'Hypertext defined and applied', Online, May, pp. 37-49.

Gordon, S.; Lewis, V. 1992 'Enhancing hypertext documents to support learning from text', Technical Communication 39(2) pp. 305-308.

Lee, N.S. 1990 'Infostation: a multimedia access system for library automation', Electronic Library, 8(6) pp. 415-420.

Moline, J. 1991 'The user interface: a hypertext model linking art objects and related information', in: M. Dillon, ed. Interfaces for Information Retrieval and Online Systems, Greenwood, N.Y., pp. 133-148.

Ragusa, J.M. et al 1993 'Adding knowledge-assistance to PC-based photographic image database management systems', Information Resources Management Journal 6(2) pp. 27-40.

Seloff, G.A. 1990 'Automated access to the NASA-JSC image archives' Library Trends, 38(4) pp. 682-696.

Willie, S. et al 1993 'Experiences with multimedia capture and storage', Significant Bits, March, pp. 33-38.

Back to Conference Papers
Back to home page

Last updated 9 March 1995