Shrinking Image File Sizes Proffers Virtual Bandwidth

April 2000
By Maryann Lawlor
E-mail About the Author

Software turns gigabytes into megabytes, allows users to view kilobytes at a time.

As militaries, governments and businesses continue to struggle with the obstacles posed by bandwidth limitations, scientists in industry and research laboratories are improving compression technologies to allow high-quality images and text to be sent to the desktop—or palmtop—with phenomenal speed. The proposition is simple: Until scientists design a way to make the communications pipelines larger, engineers must make the volume of data smaller.

Improvements in telecommunications capabilities have opened up numerous opportunities to anyone with information to disseminate. Although early Internet users employed the technology for collaboration by sending relatively small files of research data, today’s demand to send audio, video and interactive material to millions of viewers has researchers exploring ways to cope with a burgeoning market.

Bandwidth challenges cross the spectrum—from military commands that want to share critical satellite images to companies that yearn to deliver lifelike pictures of their products in an effort to lure buyers. Federal, state and local governments are in the process of pouring more service-to-the-citizen programs onto the World Wide Web to facilitate outreach efforts. The systems that these activities have in common come in various shapes and sizes, but their users all must deal with the same problem—delays. Online buyers are put off by the wait for Web sites to download, and on the battlefield, delays in transmitting data can influence the outcome of battles.

Several companies and research facilities are taking different approaches to designing technologies that compress the information at the sender’s end—whether it is satellite images, photographs or text—so that it can move across the connections quickly and then be decompressed by the receiver. These techniques increase the quality of the data that can be viewed while at the same time escalate the speed at which it can travel over existing wires.

Software developed by LizardTech Incorporated, Seattle, Washington, dramatically reduces the number of bytes required to capture a useable image. Unlike other compression technologies such as jpeg, which offers approximately a 4-to-1 compression ratio before introducing blocking artifacts, the company’s Multiresolution Seamless Image Database (MrSID) software delivers visually lossless images that are compressed at up to a 30-to-1 ratio. MrSID can take an 80-megabyte, high-resolution image and shrink it 97 percent to a 2.4-megabyte reproduction-quality, single-source image. Once encoded into a MrSID portable image format, the MrSID Image Server software for Web servers can display the image instantaneously for e-shoppers. In addition, selective decompression software facilitates image viewing by decompressing and loading only the section of the image that is currently being seen on the monitor.

Compression techniques are especially useful to aerial and geospatial information systems (GIS) customers. Thomas Foley, geospatial product marketing manager, LizardTech, offers an example of the amount of compression MrSID offers its users. “In aerial photography, if you have an image of Washington, D.C., every pixel represents about 2 feet of ground space. When the image is originally put into digital format, it would require about 8 gigabytes of memory. We translate, or encode, it into a MrSID file, and during that procedure we go through the compression. MrSID compresses it to [a ratio of] 24-to-1, and we end up with a file that is 400 or 500 megabytes. Then, through selective decompression, the viewer is not loading all 400 megabytes at one time but probably only loading a couple of hundred kilobytes at any given time. That gives you the speed,” he explains.

The origin of the technology is the Los Alamos National Laboratory, New Mexico, team that also developed the current Federal Bureau of Investigation standard for fingerprint image compression. LizardTech was formed in 1992 when it received the initial technology license. After creating several products for the Macintosh market and receiving recognition for its work, the firm began designing an industrial-strength image compression technology. MrSID was launched in 1996, and according to Mark Weeks, product manager for MrSID, it is now the de facto standard in the GIS industry.

The software is based on discrete wavelet transform. “MrSID takes the image and breaks it down into frequencies—high frequency and low frequency—through a process called wavelet decomposition,” Weeks explains. “The next step, based on the compression ratio desired by the user, compresses the imagery. If the user selects a ratio that is visually lossless, MrSID maintains only the image data that the human eye can see and throws away what the eye can’t see. A good analogy would be a suitcase where you unpack everything then repack it so you can fit more stuff in it,” he adds.

“At the same time, the MrSID file structure allows for multiresolution imagery. It enables selective recompression and creates seamless mosaics, or in other words takes multiple tiles and puts them in one seamless file. To view the same image in jpeg or tiff format, you would have to open all the tiles. With MrSID’s selective decompression, it decompresses the image and opens up only what the viewer wants to see. You can effectively open all of the tiles for viewing simultaneously with little overhead,” he adds.

In contrast, fractal-based technologies are not mathematically accurate. They could be throwing away things the viewer would need, Weeks says. “Additionally, jpeg uses discrete cosign form. You can only achieve a certain amount of compression with it due to jpeg’s blocking artifacts. In general, it does a 4-to-1 compression, where MrSID can do a 20- or 30-to-1 that is visually lossless,” Weeks offers.

From a telecommunications standpoint, these capabilities make the transmission of images “virtually painless,” Foley says. “In terms of telecommunications, meaning bandwidth and the Internet and all those kinds of things … MrSID can offer higher quality. People can’t make a buying decision right now because of the quality [of the images]. If companies can present a high enough quality, from a marketing standpoint, it’s a solution,” he adds.

The federal government also has realized the benefits of compression technology. The Library of Congress was Lizardtech’s first customer for MrSID. After incorporating the software into its Web site, more people viewed the available maps in one week than they had in 11 years because of accessibility, Weeks says.

From a military standpoint, the capability has been similarly beneficial. According to Foley, the technology offers the same advantages to the armed forces as it provides to the consumer, and the company is working with every branch of the service. In the past, satellite images would be saved on a disk or CD that would have to be physically transported to a location. When formatted in MrSID, these images can be sent directly to the field quickly. This month, the firm is scheduled to release a version of the software that can be embedded in another company’s product and will allow images to be opened on Windows CE, so information could be available for decision making from a handheld platform. Although a near-real-time capability is not yet possible, it is not out of the question, Weeks says. To this end, company researchers are examining wireless transmission. In addition, although security technology currently is built into the encoder piece, engineers also are investigating how to make the security product modular so that military or other customers could plug it into whatever device they are using.

Governments are one of the company’s biggest customers, including jurisdictions at the state, city and county levels that use geospatial images for city planning or to assess damage caused by natural disasters.

Photographs are not the only items that can be compressed, and LizardTech is not the only company addressing the bandwidth limitation challenge through compression techniques. While LizardTech focuses on technologies that shrink files such as GIS images, DjVu, developed by AT&T Labs Research, Florham Park, New Jersey, allows the Internet distribution of high-resolution versions of scanned pages and compact versions of digital documents. For example, a document that is 31.2 megabytes in a tiff format or 604 kilobytes in a jpeg format can be reduced to 70 kilobytes in DjVu.

The product combines six technologies from AT&T Labs, including a two-stage decompression algorithm for on-the-fly, on-demand image decompression; a foreground-background separation technique; an IW44 wavelet-based progressive image compression technology; JB2 bilevel image compression technology; a ZP-coder binary adaptive entropy coder; and a projection-image masking technique. The document image compression technique achieves compression ratios as high as 1,000-to-1. Scanned pages at 300 dots per inch in full color can be compressed from 25 megabytes to between 30 and 80 kilobytes.

Like MrSID, DjVu does not decompress an entire message at one time but rather keeps the information in memory in a compact form, then decompresses it as each piece is displayed on the screen.

Richard V. Cox, division manager, AT&T Labs Research, explains that many of the methods that are used in data and image compression can also be applied to audio and video information. “To differentiate a bit in the image, you have to work with what the eye can see. You keep that and throw away what the eye can’t see. The same technique applies to audio coding. In audio coding, you keep what you can hear and throw away the rest. How many bits do you need to describe the audio? Then you can throw away the parts that are masked,” he explains.

According to Cox, compressed speech technologies were first used during World War II; however, these initial approaches offered a limited output quality and still required a great deal of computational hardware. Today’s digital wireless communications services rely on compression technologies. “Without the compression, far more bandwidth would be needed to support the current number of wireless subscribers,” he says.

This also is the case with video, particularly with the introduction of high-definition television (HDTV). Regular television is analog. “If it were digitized, it would be 30 frames per second, 525 lines and typically about 750 picture elements per line. To describe each picture element takes 24 bits. In uncompressed form, this amounts to a bit rate of roughly 250 million bits per second. Satellite and digital cable are using between two and three million bits per second. That’s about a 100-to-1 compression. With HDTV, you have more frames per second, more lines per second, more bits per line, and with compression you can get down to 10 megabits per second,” he explains.

These compression numbers now need to be achieved in transmitting video to computers for video to the desktop to be a viable option, Cox says. “As we evolve toward streaming video to a computer, we need to balance the video bit rate and the communication bandwidth. On 10-megabit-per-second Ethernet, 2 to 3 megabits per second of streaming video is too much. It will clog the channel for other users. If the channel bandwidth of the Ethernet is 100 or 1,000 times greater, then 2 to 3 megabits per second fits easily,” he explains.

Although fiber optic cabling is one solution to this problem, current router technologies are not yet capable of handling the speed that fiber provides. “So, you have the fiber, but you can’t fully use it until the rest of the technology catches up. You’ll see it more at home very quickly between digital cable and digital subscriber line. You will see it faster if you’re willing to pay for it. These are obviously things that appeal to the higher-end users. But what is it you get for it? You get a tremendous amount,” he adds.

Both video and audio transmissions are headed toward greater fidelity and quality, and this will be accomplished through improvements in a combination of compression and recording techniques. “The whole area of video and where they’re going today is actually doing the compression within individual objects within the scene and how you restructure the scene. So the office or room where you’re sitting stays the same, and you only have to compress it once. But the things that are changing, like the expression on your face, you compress,” Cox says.

Consumers, both government and commercial, do not usually see the AT&T Labs name on products they purchase because the organization focuses on research and development, then works with a manufacturer to incorporate the technology into a finished item. “The group develops the key technology and then puts it into a standard. We want to operate a network, but we’re not a manufacturer, so we want to use it and allow other companies to use our technology to develop the actual products, and then we buy the products. The customers don’t know that it is AT&T technology that actually makes a product possible,” he offers.

“Fax, wireless, television and audio have all been using digital compression for some time. They rely on the availability of inexpensive low-power computation devices with ever-expanding capabilities. To satisfy higher demand and to create new services, compression will continue to be used even though communications bandwidth is increasing,” Cox concludes.

Enjoyed this article? SUBSCRIBE NOW to keep the content flowing.