Since the emergence of XML in early 1998 and it's subsequent adoption across diverse application domains, one of the key benefits it enabled was the separation of content and presentation [Bos97]. XML borrowed this model (along with other important concepts) from the Standard Generalised Markup Language (SGML). An SGML document consists of logically structured content and uses a separate file (style sheet) to specify how the content should be formatted for presentation. For this purpose, the Document Style Semantics and Specification Language (DSSSL), was developed. It formats SGML documents for presentation/printing, but can also transform the physical structure of the document. XML documents can be transformed and formatted in a similar manner using the extensible Stylesheet Language (XSLT), a style sheet language developed specifically for XML. In addition to transformation and formatting, other layers have emerged (and been defined through formal specification) to complete the conceptual reference architecture for XML publishing. These additional layers are generally grouped into schema (Document Type Definition (DTD), XML Schema, RELAX-NG), logic (XSP [Apa02]) and semantic layers (RDF [RDF00], TopicMaps [XTM01]). In the electronic publishing domain, the advantage of separating content from transformation and presentation information has proven very successful, particularly where content must be customised and delivered in many output formats (e.g. to suit different devices and applications). Analogous to database design normalisation techniques, final document representations were "normalised" into their basic architectural components (content, transformation, presentation). Figure 1 shows relationships and typical cardinalities between the individual components using the entity-relationship syntax.

Figure 1. ePublishing Components

The four-level publication architecture Table 1 illustrates the publishing workflow that occurs in order to generate a final document:

Figure 2. ePublishing tiers

The most widely used graphics formats on the Web are based on raster technology, where an image is composed of a flat two-dimensional grid of pixels. A pixel is the basic unit of programmable colour on a computer display or in a computer image. Graphics packages often allow images to be assembled from various graphical objects (e.g. geometrical shapes, text, other images) grouped on virtual layers. Saving the image as a raster image flattens the graphical objects into a grid of pixels removing whatever logical structures may be present. At this point, making minor changes to the image such as changing a piece of text becomes very difficult.

Vector Graphics offer a more flexible alternative to raster graphics whereby an image is composed of graphical objects (e.g. geometrical shapes, text etc.) each of which can be individually selected and/or modified. Furthermore, a saved vector graphic preserves its logical (object-) structure rather than flattening it. There is no increase in complexity between modifying a vector image at creation time or at a later stage. As a result, maintaining a vector graphic image is generally far easier than maintaining a raster graphic image. Macromedia Flash is one of the most popular vector graphic formats available and is on widespread use on the Web.

In 2001, the Scalable Vector Graphics (SVG) Specification was published by the W3C. This specification defines a vector graphic format using XML syntax that can be used to produce graphics for the Web and beyond. Like other content formats based on XML, SVG is non-proprietary, platform independent and interoperable with any existing XML aware software and tools. SVG is an important development in Web graphics and may prove to be as significant in the graphics domain as SGML and XML have been in the publishing domain.

Figure 3. Presentational Varieties

Analysis of the logical image publishing architecture exposes equivalent layers to the document publication model. However, the tasks and properties of each distinct layer are not as obvious as they are for textual content. Irrespective of the final graphics format, the following four tiers (with connecting interfaces) can be identified:

This paper outlines two adaptive graphics categories, each with distinct characteristics. On the one hand, graphics with qualitative changes are those, which will remain valid and useful for long periods of time without being changed. Modifications are only required to adapt the content of the illustration (and here mostly textual content) to local conventions (localisation). They often occur in combination with localisation of documents. On the other hand, graphics with quantitative changes are defined as images with parts that undergo changes on a regular basis. Pictures that require both qualitative and quantitative changes are also likely and defined as having mixed changes.

In all examples, we detect components of the image that can be grouped into static and dynamic layers. The static parts remain unchanged in each version of the graphic whereas the dynamic part may change according to the users needs and requests.

Adaptive graphics technology physically separates the graphical and the descriptive elements within an image. Graphical elements include shapes such as circles and squares and their relation to each other. One possible format for representing graphical elements is SVG. Depending on the requester's properties, the SVG file is processed and completed with descriptive information and transferred in SVG format to the client.

In this case, the graphical part of the image (e.g. the layout of the rooms in a conference centre) is static and the most persistent part of the image. It is stored separately from the event specific data, e.g. occupation of rooms, title of presentations. On request, both files are merged and transformed into the final, data driven and customised image (adaptive graphic).

Another important area of adaptive graphics is the localisation domain of technical diagrams. Having separated the graphical from the descriptive elements, diagrams from domains such as the automotive, aeronautical or medical domain can easily be fitted with textual and graphical information in the reader's native language and according to their culture.

It is widely accepted that separation of content and presentation information within the area of electronic publishing is important. Equally we have learned that normalisation of graphical content can yield similar benefits for the creator and user as well as the person responsible for maintaining the content. SVG is an important technology and is well suited to the process of customising and adapting graphics to users' needs. It has the potential to be as liberating for the graphics community as XML has been for the publishing community.

Should we also consider normalising the remaining multimedia formats, i.e. audio and video? What advantages can we achieve by doing so? Does the overhead involved in creating the necessary metadata outweigh the possible advantages? Can we use existing specifications such as Synchronized Multimedia Integration Language (SMIL)[SMI01] and HyTime [HyT97] as a basis for this? These are interesting questions, complex to specify and implement.