IS&T/ SPIE
Symposium on Electronic Imaging: Science and Technology (EI'99)IS&T/ SPIE
Symposium on Electronic Imaging: Science and Technology (EI'99)We are keeping the panel description as broad as possible so we can decide in the last minute the details of the discussion. Here is a short description of the event:
A few years after Gutenberg invented the moveable type as an enabling technology, Manutius created the revolutionary publishing business, one of the main propellers of the Renaissance. When Xerox PARC invented the workstation, it hailed the paperless office. Yet, after over 20 years we have not seen a fundamental revolution as Manutius brought forth at his time.
The enabling technology has improved from the Alto, to PageMaker, to the Docutech, to the World Wide Web. PageMaker allows authors to do their own pre-press work and send it directly to press, bypassing the graphic arts industry. The Docutech allows printing on demand, eliminating the distribution and warehousing problems. The World Wide Web allows capitalizing on specific knowledge, which is much more valuable for competitive advantage than the general knowledge capitalization introduced by Manutius.
In this panel discussion eminent representatives from academia and industry will debate on possible technological barriers in electronic imaging that might cause the hold up of the electronic publishing revolution. This panel is a continuation of two successful debates: 1997 on Internet printing, and 1998 on the future of electronic publishing.
The title came out of a quick brainstorm by the panel members and some active people in the audience in January 1998 at the conclusion of that panel. The criteria were:
The way we handle the panel is that the first week of January I will discuss a possible theme with John Michaelis. John will then flesh it out and send e-mail to the panel members and solicit their opinion. He will then send out a summary.
On Wednesday January 27 John will hold a lunch (Dutch) for the panel members in a restaurant at walking distance from the conference center. This lunch will serve for introductions and to strategize on the panel discussion.
There are no slides, etc. each member will have time for an initial statement. Then there is discussion among the panel members with participation from the audience. Usually the audience is very active. The members of the panel are very experienced, so it will be a very intense event.
It is time to reconsider our research priorities. The veterans in our community started out in an age when the image processing tools in the publishing industry were graphic arts cameras and screens, pin registers, knifes, and goldenrod paper. Their contribution was to invent the technologies that brought us from mechanical processes to electronic publishing: scanners, image enhancement, colorimetric color control, color management, perceptually lossless image compression, and digital halftoning among others.
It is a sign of the exponential progress in science and technology, that the same generation of scientists and engineers is able to contribute to a new paradigm shift, namely from electronic publishing to digital publishing.
Digital publishing is a fully digital process, from end to end: cameras, scanners, processing, storage, syndication, distribution, and rendering. Many underlying electronic imaging technologies remain the same. However, the new fully digital process changes the emphasis of which goals are more important, i.e., it requires us to reconsider the priorities in our research programs and objectives. We briefly review the impact on some application areas. One example is printer resolution.
Digital cameras are now generally accepted as the most appropriate tool both in the studio-e.g., for catalog work-and in the field-e.g., sports and general reportage. Today's professional digital cameras fulfill the needs of their users and the main challenge for electronic imaging is to reduce the device cost by an order of magnitude to bring it more in line with the cost of the old AgX-based technology.
For the amateur market, we must revise our way of thinking. With the AgX technology, an amateur application is a scaled-down and simplified version of the professional application that can achieve more or less the same image quality. Over the last decade the photo amateur has been replaced by the photo consumer, and any digital photography application for the consumer must compete with disposable cameras. In the last couple of years we have seen the emergence of a new paradigm for consumer photography, namely the instantaneous and casual communication of candid images.
In this new paradigm for consumer photography, digital cameras are peripherals for hand-held computers and have to obey the rules for such peripherals, namely be very small and cost as much as a mouse. Regardless of these two constraints, the digital consumer camera has to work anywhere-like a disposable camera-but without a flash, because batteries are too bulky. These requirements place emphasis on two electronic imaging research topics: dynamic range and color constancy.
Dynamic range does not necessarily mean more bits per pixel, which are probably not possible under the cost and size constraints. Instead, imaging algorithms such as retinex, which compress the dynamic range, have to be reconsidered. Similarly, incomplete adaptation in available light situations call for pleasing and not too harsh color rendering algorithms based on the judicious application of color appearance models.
Contrary to the spartan requirements for digital cameras, image processing benefits from a bonanza in desktop computer power. With 400 MHz processors on 100 MHz system busses and 128M bytes of RAM, today's personal computers have bandwidth to spare after the user's primary requirements have been fulfilled. This leaves considerable performance available for improved color imaging.
New algorithms, more heavily based on non-linear methods and operating in more suitable representational spaces, can now be deployed even in such performance-critical system components as operating systems and device drivers. For example, if the optical properties for an inexpensive simple input device are well characterized, sophisticated algorithms can be used to restore images to unprecedented quality.
On August 13, 1997, U.S. District Judge Sonia Sotomayor rendered a very important decision in Tasini et al. vs. New York Times et al. regarding copyrights (see http://www.nwu.org/nwu/tvt/tvtrule.htm). Her decision, based on section 201 (c) of the Copyright Act of 1976, which deals with the copyrights in collective works, reads that while the publisher of a collective work retains the copyrights for further publication in databases and CD-ROMs-she declared them revisions of the original work- authors retain the rights to individual contributions and may license them to World Wide Web publishers, without permission from or payment to the publisher. This is an opportunity for image syndication and will spur the need for electronic imaging technology.
With perfect timing, electronic imaging researchers are delivering the appropriate technology-watermarking. However, if we consider the importance from the perspective of digital publishing, we recognize that watermarking may be a relatively small and transient problem. Because stock agencies have been early adopters of electronic imaging, it appeared that the protection of the author's rights was a major priority. This is not necessarily the case. The purpose of watermarking is the prosecution of thieves in case of unauthorized use of an image; the courts of law do not require the presence of a secret mark to convict a thief. For business purposes it is much more useful to have an standardized ownership field in the header of all image file formats, because this facilitates the clearance of copyright royalties. Almost all digital publishers are honest and research should not focus on a few malevolent exceptions.
A pressing problem that is attracting the attention of research in electronic imaging is the digitalization of the human cultural heritage. The documents created since the inception of digital publishing are only a minuscule fraction of the patrimony stored in archives. Originals are subject to various forms of decay and there is a certain urgency to digitize a huge number of documents. As much of the original information has to be preserved as possible, including pentimenti and documents under recoated media. Some of these documents are in a bad state of disrepair and can be restored only in a digital form, because the originals are too frail.
The sheer quantity of documents calls for fast electronic imaging methods, that completely capture a document in a minute at most. Complete means that an RGB scan is often not sufficient and multispectral color reproduction is necessary. When the media is not paper, geometric appearance attributes must be recorded, as is the case for silk media or oil paintings. To remove stains, soot, mildew, and other common contaminations-not to forget that acid-free paper has come in wide use only recently-sophisticated image processing algorithms are required do "dry-clean" the documents. When documents have been fragmented or torn, algorithms are necessary to stitch images and remove tear lines.
The digital archival copy of a document can have large storage requirements. This is an impediment when the cultural heritage is made public on the World Wide Web. We need algorithms that can automatically abstract that portion of the image file that is useful to the viewer at the other end of the communications line, and this requirement may be more difficult to solve than watermarking.
Capturing an image is only part of the problem. It is further necessary to catalog the image, so he can be later retrieved by searching for criteria or by navigating the space of all images. Required algorithms encompass the fields of pattern matching, object recognition, character recognition-including old typefaces and calligraphy-and reverse graphic editors to convert bitmap representations into vector representation, which is essential for the large number of geographic maps stored in archives.
Specifying the iconography, i.e., semantics of an image, is a very difficult task that requires extensive knowledge. To be useful for retrieving images from the World Wide Web, an iconography cannot be based on haphazardly assigned index words. Cataloguers must be assisted by thesauri, taxonomies, and ontologies, which are essentially all the same artifact. These structures, also know as external intelligence, allow cataloguers and image retrievers to navigate the set of images instead of performing flat keyword searches.
Cataloguers are aided by image classifiers, which in turn can use the iconographic structures to improve their accuracy. Image classifiers are also becoming increasingly important for color gamut mapping. Depending on the output device, the rendering algorithm has to map the colors in the image to a gamut that has a considerably different gamut then the range of colors in the image. It is necessary to distort the color in the image but some fundamental memory colors such as complexion cannot be changed too much.
Images are becoming increasingly accessible because of the availability of the Internet as an inexpensive, convenient, and universal communications medium. The World Wide Web implements a graphical user interface with a hypertext linking system that facilitates navigation. Both the iconography of images and the requirements of users can be expressed in the structured language XML, allowing servers and browsers to negotiate the best rendition of an image for a particular contact. Electronic imaging provides the algorithms for this system.
We have already touched on the problem of gamut mapping. There are more problems for electronic imaging. The modulation transfer function (MTF) of today's printers is better than that of the human visual system. Concomitantly, the considerable image quality improvement in one-hour photo finishing processors has risen user expectations. Consequently, the careful design of color savvy halftoning algorithms has become much more critical and it is no longer possible to apply the same grayscale algorithm to each color plane. With increased image sharpness, color artifacts are also more visible and color matching must be more accurate.
Only a few years ago, printer MTF was so poor that images could be aggressively compressed using the example quantization tables from the JPEG standard specification. Today the quantization tables in lossy data compression algorithms must be designed very carefully to be perceptually lossless for the intended rendering devices and viewing conditions.
The increased use of images in documents poses new problems also to the choice of data compression algorithm classes for images. While JPEG and wavelets work well on full color images communicated over the Internet, these algorithms fail on halftoned images. The last cable to the printer has become a severe performance bottleneck and we need lossless compression methods like JBIG, which do not destroy the structure of halftones.
I have touched on some components of a digital publishing system, and I shall finish on the glue that holds it all together, namely workflow. As images progress from creator to viewer, the electronic imaging algorithms must be able to operate in a pipeline. There is agreement on device profiles (although the implementations are still nightmares for users), file formats, communication protocols, and markup languages. However, other issues, such as end-to-end MTF optimization, color spaces, and rendering intent, are still open.
When we design electronic imaging algorithms for digital publishing systems, we shall not forget that some of our technologies are taking a large human toll. Our new technologies are the basis for tools allowing everyone to do their digital publications by themselves. The pre-press industry, which worldwide employs an amazing number of imaging experts working in small businesses, is on the verge of disappearance. While mechanical paste-up assembly and stripping have gone the way of the buggy whip trade, we should not waste the specific knowledge in the pre-press industry. These talents are very much needed in the new world of digital publishing, especially now that we have to reconsider our research priorities. For example, a trained eye can take us a long way in end-to-end MTF optimization and choosing the most appropriate compression algorithms.