Video compression on the move

Richard Doherty
Electronic Engineering Times

March 6, 1989

The worldwide drive to deliver high-performance image processing for color video continues apace, with real-time compression the most recognized, though certainly not the only goal. Silicon houses and software vendors in the United States, Japan and Europe have an eye on the growing market for graphics and imaging, especially in areas like teleconferencing.

Chip makers are exploring silicon-based "compression engines" that deliver automatic edge detection and video-noise cancellation, along with other advanced graphics manipulations. Software engineers point to packages already up and running on PCs, Macs and workstations that digitize and compress video images, albeit not in real time, so that they can be transmitted efficiently over LANs and telephone lines or stored to disk.

Universal Video Communications Corp. took a step ahead of the pack last week when it confirmed that its VP-2000 racked up a number of industry firsts, including the ability to handle color video compression and playback in real time (see Feb. 27, page 1). UVC chose a silicon-based scheme that strips down the normally computationally intensive, memory-hungry video processing.

Critics of UVC's approach claim that it has achieved real time at the cost of the broader palette features that are typically achieved in software.

A more conventional route to video compression engines can be seen in Intel's Digital Video Interactive configuration. DVI is now in beta testing, with products slated for later in the year (see Oct. 10, page 1).

These schemes encompass one or more of the following:

From an original video source (camera, tape or disk), small clusters of pixels are transformed into a spatial domain or a frequency domain, or both. They are then evaluated on some criteria and encoded. The encoded product is represented by smaller amount of data than the original image file. To further cut down the size of To further cut down the size of the file, the data is further tweaked. For instance, every other horizontal scan line can be skipped.

Traditional color NTSC video sources have an image data rate (at 8 bits per color) of 150 million bps. At that rate, two minutes of digitized information would require a 2.25-Gbyte Winchester drive. Compression clearly squeezes the size of resultant image files, but it's easy to see why large buffer memories are generally associated with image processing.

UVC engineers decided that standard approaches were too costly in terms of calculations and that reconstruction engines needed to expand a compressed file needed too wide an information bandwidth (the encoded data plus the reconstruction math products). Instead, they use an approach that stores an equivalent two-minute NTSC color-image sequence in 9 to 54 Mbytes, depending on the image quality and desired playback frame count. That's a compression ratio of somewhere between 200:1 and 80:1.

Rick Stauffer, head of marketing at Intel's DVI group (Princeton, N.J.), said his team is addressing "a broader set of applications [than UVC!. We're keeping our eye on some of these teleconferencing people, but DVI will cater to users who need to manipulate an image on a desktop PC."

Stauffer pointed out that the DVI chip set allows users to perform 360 pans and real-time warping of image elements-features not currently in UVC's charter.

"Our position is we have a more flexible technology, and probably we're farther ahead in some key areas. We have working chips, and our custom versions will be a lot faster than [UVC's! 12.5 Mips," said Stauffer. Explaining the long road to bring DVI to the desktop, he remarked: "It's become very evident to me how complex it is to take technology from the lab to the marketplace."

Stauffer expects some of the work being done by UVC and other so-called teleconferencing companies will come down in price and complexity.

DVI comprises two image-processor chips. The first, VDP1, is rated at 12.5 Mips. VDP2 is a display-output processor. Together, the chips are able to perform combinational meshing of computer graphics, text and live video, all while maintaining varying degrees of audio digitization and playback.

When collage-style graphics (mixtures of complex computer graphics, mapped textures and digitized video images) are needed, the existing beta version of the DVI development system has no competition. However, because the compressed image data formats produced by UVC's technology are merely data files, there is no technical reason why UVC, or third parties, might not produce software tools to allow a clean combination of text, surface textures and real-world video images.

To achieve the over 100:1 average compression ability touted by DVI developers, however, the original images and audio must be preprocessed. This workstation-level compression requires even more Mips than the DVI chip set can deliver. As a result, DVI leaders have spoken of using modular Transputer-based compression systems. These work far from real time, requiring about a 90:1 ratio to squeeze image data.

According to DVI technical literature, expansion of compressed still images "even at high resolution, is accomplished in a fraction of a second." Low-resolution images are restored in real time, at rates to 1/30 of a second. When used to store images on a CD-ROM with 600-Mbyte capacity, DVI can store 7,000 768 480-pixel, high-resolution images; 10,000 medium-resolution 512 480-pixel images; and 40,000 256 240-pixel images.

Last spring, the RCA DVI team added Edit Level Video (ELV) to its PC-based development platform. This allows a single-frame image to be compressed in 1 or 2 seconds. Compression ratios of between 7:1 and 25:1 can be expected using this technology. But DVI gives the caveat that "the exact reproduction required for graphic illustrations or medical applications is just between 2:1 and 3:1."

Adding motion video compression to this technique (using DVI's specialized service) brings it close to 3 seconds per frame, 60 times that of one year ago. At that speed, a 1-hour video presentation would require a minimum of 90 hours to compress. In contrast, UVC works in real time.

Intel senior vice president David House said "a slew of DVI announcements" are slated for Microsoft's Fourth International Conference on CD-ROM in less than a month. It was at the second Microsoft conference that RCA's Sarnoff Labs unveiled DVI to the world.

DVI grew out of the video expertise RCA applied to its failed capacitive video-disk project (CED) in the early 1980s. The video-disk was a $4 billion washout, but the core research team that developed CED so believed in the power of digitally modified video that DVI was born from CED's ashes.

The DVI research effort was spared from transfer to France's Thomson as part of its acquisition of General Electric.

In October 1988, Intel purchased the DVI group, renaming it the "Intel Princeton Operation. According to the timetable announced last November, Intel's role is to help drive the commercialization of DVI "beginning in 1989." Intel is said to be reducing the cost of the two DVI imaging chips and support logic. Board-level products are expected this year, and Intel has pledged to bring out "low-cost, highly integrated DVI products based on a new Intel DVI chip set" in 1990.

DVI has racked up hundreds of commercial inquiries and over a dozen developers. Intel says the chips are destined to appear in home electronics products as well.