May 1998

BY NANCY MIRACLE

Let’s face it — digital video conferencing isn’t going to become the norm in the business world until the image on the computer monitor is as good as what people see on their home TV. In a world where television and video games abound, we have become accustomed to seeing a clear, sharp picture. Visual information is attractive and is often the most effective means of conveying extremely complex information and exact shades of meaning. Why then, hasn’t the digital video conferencing business been exploding?

Research shows that people assimilate visual information more quickly than audio or written data. The human brain is capable of transforming complex visual images into meaningful information at very high speeds. This ability may originally have had a clear evolutionary benefit (when applied, for instance, to such problems as the recognition and evasion of fast-moving predators).

Information professionals are increasingly being forced to deal with a new problem: How to deliver even more information to users who are already overwhelmed with data. Most system implementers have had the unpleasant experience of installing a system or service that "should have been" extremely useful, only to have it go unused, often for largely inexplicable reasons.

The use of digital video in business computer applications has had surprisingly little success, at least until driven by the explosion of graphics on the Internet. Although widely discussed, the deployment of multimedia data in the business environment has been significantly hampered by problems with the attractiveness of the video. This is a result of the high cost of delivering large data streams.

DEALING WITH DATA SIZE
A textual data screen of 80 columns by 24 lines, refreshed on demand approximately once per minute, generates a data stream of 256 bits per second (0.000256 Mbps). In comparison, uncompressed audio data requires approximately 64 Kbps (0.064 Mbps).

Both of these streams pale next to video data, which is extremely bulky. Simple NTSC (the U.S. commercial television standard) or PAL (the European commercial television standard) images, shown at standard display rates, generate data streams on the order of 100 Mbps. The combined data streams for a raw audio/video data stream therefore require about 400,000 times the band-width of the text data stream.

REDUCING BANDWIDTH REQUIRED
The cost of transmission increases directly with the amount of bandwidth required. Thus, one barrier to the deployment of video systems has been the amount of bandwidth that is needed to transmit real-time video.

The designer of video transmission systems is helped by the fact that video streams, although very bulky, often contain large amounts of redundant data. In the business context, much visual data is "quiet," with monochrome backgrounds. Furthermore, there is often little difference between images. In an image of a person talking, the amount of change between frames is typically limited to the area surrounding the head. Similarities both within and between images can be exploited to reduce the amount of bandwidth required for transmission. This process is called data compression (see sidebar entitled Six Steps Of Data Compression).

In addition to having different results on the final decoded image, each step requires a certain amount of computational resources. As a general rule, the more computer power that can be applied to the tasks above, the smaller the resulting data stream. It is the job of the designer of the video transmission system to select the combination of data compression steps to provide an image of the desired quality. The amount of computing power required to create the image is a trade off against the amount of bandwidth that is available during the transmission.

ADVANCES IN COMPRESSION
Initially driven by military requirements for the efficient transmission of extremely detailed data from satellite sources, research in signal processing has focused extensively on the mathematical algorithms used for sampling and compression. One of the most promising developments of the 1980s was the application of the wavelet transform and its discrete-time cousin, filter banks, providing sub-band coding to video compression.

Wavelet transforms operate on a continuous pixel stream, rather than on macro blocks. This results in a tangible difference between the algorithms that utilize discrete cosign transforms with macro blocks and those that use wavelet transforms. Because of the human ability to recognize patterns, viewers are very sensitive to defects that involve repetitive horizontal or vertical elements. Defects from algorithms that use macro blocks are strongly geometrical and are easily perceived by the user. At best, viewers find them distracting. At worst, they make the image almost impossible to view. In contrast, the continuous pixel compression technique used during wavelet compression causes these images to look soft and muted at higher compression rates, but image detail is not lost and transmission defects are less noticeable to the viewer.

Even more significantly, wavelet transforms achieve compression ratios that are impossible with other types of compression. Data can be compressed at 100:1 rates with virtually no visible artifacting, and at rates of up to 300:1 providing an accept able QCIF picture. QCIF (Quarter Common Intermediate Format) is part of the ITU-T’s H.261 standard for video conferencing.

COMPRESSION APPLICATIONS
The compression and reconstruction of satellite image data was one of the first practical applications of wavelet transforms to image processing. One characteristic of the wavelet transform is that image detail is not lost as quickly as it is lost with algorithms that operate on macro blocks. This meant that wavelet transforms were ideal for the compression of medical image data. In medical imaging, the loss of small image details is unacceptable. In 1997, the first commercial video communication system that used wavelet compression became available. Today wavelet compression is being proposed as an integral part of many of the new compression standards. The JPEG 2000 compliance committee is already examining wavelet image compression and it is expected that the MPEG committee will do likewise.

Only a few algorithms are used in video compression for commercial applications. Each has been designed for specific purpose, each uses different elements for compression, and each has characteristic performance and defects.

CHOOSING A COMPRESSION ALGORITHM
The designer’s choice of compression algorithm is significant to the information professional for a number of reasons. Although the deployment of video in business has been hampered by cost, it has been even more significantly hampered by poor video quality. Historically, businesses have deployed low-bandwidth systems that use the H.261 algorithm. Of all the commercially-used compression algorithms, this delivers the poorest quality image. Although there are some exceptions, it has been all too common for businesses to purchase expensive video conferencing systems that, after installation, went almost wholly unused. Although system complexity is often cited as an issue in user acceptance, a more common complaint is that the video quality is so poor that people simply will not use the system.

To put this into perspective, the average person in the U.S. watches approximately three hours of broadcast television per day. Thus, the user of business video systems is likely to have sophisticated expectations of the image quality and be averse to watching poor quality visual imaging. In most cases, people will (and apparently do) use audio conferencing in preference to suffering with poor quality visual data. The real benefit of wavelet compression is, therefore, that it can be used to deliver images of broadcast quality to the end user.

Virtually all the compression algorithms require hardware assistance to operate at any reasonable speed and leave any resources at all in the user’s computer. Costs of these products have been dropping rapidly. Systems that formerly cost over $50,000 to install can be purchased now for approximately $1,500. LAN-based systems are typically less expensive to install and maintain than ISDN-based systems (due to utilizing the existing network infrastructure), and IP-based systems are generally very inexpensive to operate.

The barriers to video at the desktop — poor quality and high cost — are falling rapidly. By utilizing changes in the technology that underlies video compression, the promises of video communications — attractiveness and efficiency — are finally on their way to being realized.

Nancy Miracle is vice president of operations at Intelect Visual Communications. Intelect Visual Communications Corp. (IVC) designs and markets state-of-the-art LAN and WAN video conferencing solutions. IVC offers multipoint, television- quality video conferencing products for desktop and executive systems. IVC products operate on any network that transports IP and run on Windows 95 and Windows NT. IVC is based in New York and can be reached at 800- 922-3433 .