In this paper we will discuss the common sources of electronically produced illustration material. This is not intended to be a complete analysis of commercially available systems. It is an attempt to characterize these systems for the potential designer or expert user of software for illustrations. This analysis will focus on four areas: the user model, computer graphics techniques, hardware and system requirements, and graphic arts quality.

Graphic arts quality means that we want to judge the output of the system by the existing standards of the graphic arts industry. The image should be interesting independent of its origin as a digitally produced illustration. This means not having unintended jaggies, having sophisticated, well-produced fonts, smooth lines in a variety of weights with smooth or mitered corners as appropriate, elegant arrowheads and symbols, a tasteful use of colors and textures, and commercial quality reproduction of continuous tone images. Page description languages have been designed to describe two dimensional illustrations such that these quality issues can be addressed on a wide range of devices. Our analysis will relate the semantics of the image description presented by the illustration system to that provided by a page description language.

Our analysis groups commercially available systems used for illustration into five areas: bitmap/pixel painting, drafting/CAD (computer aided design) systems, geometric illustration systems, business graphics, and full color prepress systems. These areas are not exclusive; some products span multiple areas and the trend towards integrated software systems is blurring dis-tinctions even more. Our taxonomy illuminates, however, the relative strengths and weaknesses of the different paradigms for producing illustrations.

This course is concerned with documentation, so the only pictures that are interesting here are those that can be included in a document. The minimum requirement for including the output of a picture making system in a document is that the image must be expressible in a stand-alone manner. Page description languages provide the ideal representation for this purpose. When creating images with an interactive computer system, a primary consideration is whether it is possible to aesthetically translate the screen image to the print media. The keyword here is aesthetically. Once on a printed page, we adapt the quality definitions of the print media, whose characteristics are very different from the display media's. Text and shapes rendered on a printed page at screen resolution look unacceptably jaggy. Textures suitable for displays look coarse on paper. Printers typically have a more limited number of gray levels and colors. Many vivid monitor colors cannot be reproduced on reflective media at all and approximating these colors may dramatically change the impact of the picture.

The minimum page layout system applies a simple paste-up model to positioning figures. The figures need only be reproducible, though it is helpful if each includes its own size (bounding area). To meet layout requirements, it may be necessary to scale or crop the image (in computer graphics terms: translation, scaling, and clipping). The next step up in layout flexibility is to be able to replace or remove elements. For example, commercial layout systems that accept CAD produced figures often include a way to replace the plotter-oriented text produced by the CAD system with typeset labels. The most flexible layout system would allow full editing of illustrations in place on the page. Commercially available systems do combine formatted text and graphics [9] but with limitations on the complexity of the final page image.

The combined illustrations in a document should all be stylistically similar and coordinated with the text. Style elements include text and symbol fonts, line weights, color, and texture. Line weights, dash styles, and arrowheads should be the same from figure to figure and should harmonize with the style of the document. Color coding should be uniform, color schemes should not clash and should be suitable for the final output device. It has been suggested [2] that these style elements could be separated from the geometric or schematic definition of the figures to enforce consistent style and allow for uniform media dependent adjustments. Such a separation would make it possible to combine illustrations from disparate sources by adjusting only the style parameters. However, to our knowledge, no commercially available systems support the notion of graphical style.

The definition of the ideal integrated text and graphics system for publication applications is still a research area. It must combine all the best features of graphics and text editors with a layout editor, a database for reference libraries, clip art, and revision history, be extensible to include new features as required, and have a well designed user interface, suitable for the publication professional.

As stated in the introduction, these notes will focus on four general areas when describing illustration sources. The first of these will be the user model, the model the system presents for building up a picture. Systems typically treat the image as either an array of pixels or as a collection of geometric shapes. The user can manipulate this model with a set of interactive tools. The second area will be to list some of the computer graphics techniques that are part of implementing a particular type of system. Our list will not be exhaustive but should include most of the techniques specific to the type of system under discussion, particularly those that may not obviously be difficult to implement. The third area will be a general discussion of the hardware and system requirements for a particular application. This discussion is meant to be relative rather than absolute. For example, most geometric systems require more computing power than do painting systems. Finally, we will discuss how suitable the system is for meeting the graphic arts quality issues discussed in the introduction.

There are some basic issues that are common to all the systems analysed so we will discuss them here. These topics are resolution, gray-scale and color reproduction, and graphic arts quality typography.

The output devices commonly used in preparing documentation graphics are monitors and digital printers. Resolution is a measure of the number and/or density of picture elements in a raster device. A resolution independent representation of an illustration is one that makes it possible to render an image at any resolution, achieving the best possible result for that resolution. The easiest way to achieve resolution independence is to use a geometric representation which can be smoothly scan converted to an arbitrary resolution as the basis for the illustration. Pixel oriented illustrations are difficult to translate to paper without visible aliasing effects or jaggies.

Normally, an illustration will be designed on a monitor and ultimately output to a printer. A typical color monitor resolution is 640 by 480 pixels and each pixel can have up to 256 intensity levels for each primary color. Black and white displays are often higher resolution, but still on the order of 1000 pixels across. The spacing of the pixels depends on the size of the monitor tube, but 72 spots to the inch (spi) is a typical resolution for a black and white display. Digital printers have a wide range of resolutions, up to 2000 spots to the inch, but they are bi-level devices. Typical resolutions for dot matrix printers are around 100 spi or less, for laser printers around 300 spi, for photographic image setters about 1000 spi, and for prepress quality film plotters up to 2000 spi.

In summary, monitors are low-resolution but may have intensity levels whereas graphic arts quality printers are high-resolution but must simulate intensity levels with patterns modeled after the halftoning techniques developed for the offset printing industry. Line art must be anti-aliased to appear smooth on a monitor, but will appear smooth at resolutions of 300 spi and greater when printed. Text must be specially crafted to be readable at monitor resolutions but can be rendered from outline representations (with care) at 300 spi and easily at 600 spi and up. In general, higher resolution devices produce higher quality images and cost more. Illustration systems must balance the desire for high-resolution with the overall cost of the system.

Intensity variations on a color monitor are achieved by exciting the monitor phosphor to produce different amounts of light. Intensity variation on a digital printer is achieved by the use of patterns called halftones, as in conventional offset printing [8,17]. The technique of halftoning trades spatial resolution for intensity levels as shown in Figure 1. The number of gray levels obtained by this technique is a function of the number of dots simulating the halftone pattern. Since the printer dot size is fixed, there is a tradeoff between edge sharpness and the number of gray levels. Commercial printing typically uses halftone frequencies of 60 to 150 lines per inch. To simulate this on a digital printer and maintain approximately 100 gray levels, we need printer resolutions of approximately 600 to 2000 lines per inch.

One common problem with printing computer generated continuous tone images is the appearance of contour lines in smoothly shaded areas. These contour lines are caused by quantization in the printed rendering of shades of gray. This is caused by the limits of the resolution available to simulate the halftone screen. For example, using a 4 by 4 array of pixels to model a halftone dot produces a maximum of 17 gray levels. This is not adequate to produce the appearance of continuously changing shades of gray. This dramatic a limitation would cause artifacts in any example, but the problem exists even at higher resolutions. The problem is worse for computer generated images than for scanned images because scanned images have background noise level as an artifact of the scanning process which tends to mask the contours whereas computer generated images are noise free. One solution is to add random noise to the image to break up the contour patterns.

A color monitor produces colors when the electron beam stimulates red, green, and blue phosphor dots to produce a pattern of red, green, and blue luminous spots. The brightness of the dots is a function of the power supplied by the electron beam. The color system described is additive and the three primaries are independent. A color for a monitor, therefore, is specified as a red, green, blue triple, hereafter called an RGB value.

A color printer produces color by overlaying spots of magenta, cyan, yellow, and (optionally) black inks on paper. Patterning inks on paper is an additive system, with the white light from the uncovered part of the paper adding to the colored light reflecting from the inked portion of the paper. Overlaying inks on paper is a subtractive color system, that is, the inks act as a set of filters so the color is determined by multiplying the transmission values of the inks together. Halftoning is a combination of these two color models. Mapping between the additive color model of a color monitor and the complex, non-linear color model of color printing is a difficult task. Some work in this area has been published recently [14,16]. A more general discussion of color in the digital graphic arts environment is available in a recent article [15], which has been reproduced elsewhere in these notes.

The problem of producing graphic arts quality text is discussed in detail elsewhere in these notes [3]. It is sufficient to note here that fonts designed as rasters for displays will look terrible on paper. Furthermore, any small shape or symbol, such as an arrowhead, has the same rendering and representation problems as fonts.

In summary, the most significant factor in analyzing illustration systems is the underlying representation presented by the system. CAD systems and geometric illustration systems work on a geometric model of the image. Painting systems and digital prepress systems use a raster representation of the image. Business graphic systems may use either, but are more likely to be geometric in the design of the initial diagram.

Systems with similar representations tend to share user interface concepts and will tend to merge in the future. For example, prepress systems are importing many of the techniques common to painting systems. Painting systems are using low-cost video cameras to add scanned images to painted illustrations. CAD systems are being implemented on low-cost personal computers and promoted as technical illustration systems. Research in geometric design systems is exploring how to adopt precision placement techniques common to CAD systems [4] for illustration design. There is a trend in business graphics systems to allow the user to customize the illustration using a painting system. This operation destroys whatever structure was present in the illustration. Hopefully, as the market develops a demand for higher quality business graphics, this trend will be redirected towards merging business graphics with geometric design systems.

Raster oriented systems are more flexible and usually are more attractive to users, whereas geometric systems produce higher quality output and potentially have more power. In the long term, it should be possible to combine painting techniques with structured techniques in a single system to produce a significantly enhanced set of functions for producing illustrations.