Confused about compression? Here's how the leading codecs stack up for speed and image quality.
John Bryan
Video compression is like life: It's impossible to be too thin or too rich. In the video world, crash diets and wealth take the form of new algorithms that reduce video to its essence but keep it looking as rich as the original image. Packing video data tighter means it will look more realistic when its played back on affordable video hardware, or better yet, on the same hardware that's sitting on our desktops crunching data and processing words.
A number of compression schemes exist, each with its strengths and weaknesses. For example, just because Radius's Cinepak shines for high-motion video clips, don't expect it to be tops for a low-motion "talking-head" clip; image quality is sharp a
nd few codecs (compressor/decompressors) work better than the venerable Video 1 when images change little from frame to frame.
Video compression is vital to anyone using digital video, whether you're a developer of CD-ROM titles and videoconferencing systems or a creator of corporate presentations and training tools. The codecs you choose for each project will determine how good the images look and how smoothly they flow when your audience watches them.
Compression Camps
Compression methods fall into camps. Some rely entirely on software codecs, and others use a combination of software and dedicated processors. Hardware-assisted compression algorithms have one terrific advantage over the software approach: speed. Plug in an MPEG board, and you can decompress video at the same rate and image quality as that of a VCR tape. Besides high cost, the downside to hardware-assisted compression is you gamble that all the potential viewers of your MPEG-compressed video will have th
e proper MPEG decoder on his or her machine.
What's the benefit of the software approach? Cost. Some leading software codecs come bundled in products like Microsoft's Video for Windows, or you can download them from a BBS. Also, the performance inferiority of software codecs is fading as Pentiums, Power Macs, and local-bus systems become the norm. The easy distribution of software codecs increases the chances that your intended audience will be able to view your video product.
How It's Done
Most compression algorithms divide a video frame into blocks and then look for redundant data. This is known as intraframe analysis. In interframe analysis, algorithms make the same kinds of comparisons between frames. The algorithms then drop redundant data to reduce the overall size of the video file. This technique is called lossy compression, and it's used by all the major video codecs. In general, the more you compress a frame or data stream, the more losses there are. The trick
is to balance the compression ratio with the resulting image quality.
Codecs may be either symmetric or asymmetric. A symmetric algorithm uses an equal amount of time for compression and decompression and is common in real-time video capture in such applications as videoconferencing. Asymmetric codecs take more time in the compression stage to try for the highest level of belt tightening possible and work best for presentations or CD-ROMs.
Once a codec saves the nonredundant data, it statistically processes the arrangement of pixels in an image. These statistical data arrangements become encoded with an RLL encoding scheme or an equivalent, with shorter codes assigned to repetitive data.
The codec then may use a variety of compression methods. Two popular ones are DCT (discrete cosine transform) and VQ (vector quantization). Both use wave equations or a similar mathematical concept. The frame is divided into blocks, generally 8 by 8 pixels each, and the transform mechanisms work on the in
dividual blocks.
DCT converts pixel intensities to a frequency-based equivalent. The transform yield is a series of numbers that represent ever-finer detail in a pixel block. Compression comes by eliminating the representative numbers after a certain point in the series. The result is a loss of fine detail, but ideally that level of detail won't be detectable. MPEG codecs use DCT.
VQ uses vectors to select predefined equations. Codecs can tune quantization so that block color and intensity are handled differently on a discreet basis and in relationship to the surrounding image values. Both Cinepak and Indeo use VQ for intraframe compression. One benefit of the VQ algorithm is that it is based on a table comparison process to provide fast performance on high-end 486s and Pentiums.
Other less widely used compression methods exist beyond DCT and VQ, including fractal-image compression, a proprietary technology developed by Integrated Systems' Michael Barnsley (for details, see "Fractal Image
Compression," October 1993 BYTE). An enhancement of the still-image JPEG standard, M-JPEG (motion JPEG) compresses every frame individually for high image quality.
The result of the transform process is then compressed via RLE (run-length encoding). Codecs compress these bit strings even further by scanning the patterns at a different angle and converting the values to RLA (run length/amplitude) tokens. Huffman coding then converts the most frequently occurring tokens into the shortest bit strings.
Codecs use these compression transforms for intraframe compression. Interframe compression is largely a subtractive process that a general-purpose CPU can handle.
Microsoft and Apple, with Video for Windows and QuickTime, respectively, influence the compression world, although they don't specifically provide their own codecs. The AVI (Audio Video Interleave) format in Video for Windows provides a standard file format for developers. Video for Windows includes Cinepak, Indeo, and Microsoft's RL
E and Video 1.
QuickTime for Windows and QuickTime for the Macintosh support Indeo, Cinepak, and MPEG decompression boards. QuickTime for Windows files and Video for Windows files aren't compatible.
How to Choose?
To pick a codec, first decide if your audience is likely to have dedicated video hardware or general-business computers. Then select an appropriate resolution and color depth, depending on the subject of the video. Finally, process your video with a variety of codecs to compare image quality. With your eye as the final judge, your video files should be both thin and rich with impact.
PRODUCT INFORMATION
Cinepak Free to developers
Radius
Sunnyvale, CA
(800) 227-2795
(408) 541-6100
fax: (408) 541-6150
Indeo 3.23 Free to developers
Intel
Santa Clara, CA
(800) 538-3373*
fax: (800) 628-2283*
BBS: (9
16) 356-3600
*Accommodates international access
Truemotion S
$.50 per CD or 2 percent of revenue (whichever is higher);
$5000 prepaid royalty required.
Corporate site license of playback software: $25 per PC (100 PC minimum).
Horizons Technology, Inc.
San Diego, CA
(800) 828-3808
(619) 277-7100
fax: (619) 292-9439
Video 1
Part of Video for Windows bundle, not sold separately.
Microsoft
Redmond, WA
(800) 426-9400
(206) 882-8080
fax: (206) 936-7329
XingCD $995
Xing Technologies Corp.
Arroyo Grande, CA
(800) 294-6448
(805) 473-0145
fax: (805) 473-0147
VIEW BEFORE YOU CHOOSE
For more comparative information about codecs, consider Doceo
Publishing's Video Compression Sampler. This $99 CD offers
side-by-side playback capabilities for a number of software codecs.
The sampler includes video test files. The program keeps statistics
ab
out key frame settings, frame rates, compression quality, and
average data rate for each codec. Doceo is at (404) 876-8954.
To reflect real-world use, we compressed MPEG images at 30 fps
(frames per second) at a data rate of 150 KBps; all others were 15
fps at 290 KBps. This means the upper-right MPEG example compressed
four times more data, which accounts for the blurriness. The
lower-right box compares a software-only and an expensive
hardware-assisted MPEG codec.
Cinepak
Software-only, uses vector quantization. Originally developed by
SuperMac Technologies, now owned by Radius.
ADVANTAGES
Supports a wide range of platforms, including Windows and the
Macintosh. Part of Video for Windows and QuickTime codec libraries.
Uses a palette of 16.7 million colors.
DISADVANTAGES
Dithers to simulate colors in 8-bit mode. Host computer's pr
ocessor
and bus type determine performance.
APPLICATIONS
For high-motion video, among the best overall quality and fastest
performance, although Indeo 3.23 is only slightly slower. Not as
sharp as Indeo for low-motion video. Works best for 320- by 240-pixel
displays, quality degrades at the lower resolutions. Fuzzy images
mean it's not the best option for animated sequences.
Indeo 3.23
Intel's vector-quantization software-based descendent of DVI.
Compression technology for Intel's videoconferencing products. Can
handle DCI adapters. Version 4.0 planned for later this year.
ADVANTAGES
Palette uses 16.7 million colors. Supports Windows, Macintosh, and
IBM's Ultimedia OS/2 R2.1 and Video In. Part of Video for Windows and
QuickTime codec libraries.
DISADVANTAGES
Dithers to simulate colors in 8-bit mode. Host computer's processor
and bus type determine performance.
APPLICATIONS
Among t
he best in performance and quality for high-motion video.
Performs well on low-motion videos at virtually all resolutions.
Better palette handling in 8-bit mode than Cinepak. Indeo is best
choice when working with resolutions lower than 320- by 240-pixels.
Fuzzy animations.
MPEG
MPEG-1, MPEG-2 (currently used primarily by the broadcast industry),
and MPEG-4 (for video telephony) developed by the ISO Motion Picture
Experts Group. Based on lossy asymmetric compression algorithms.
Designed to be played back with dedicated hardware, although
software-only playback products are starting to appear.
ADVANTAGES
Produces VCR-quality video at full-motion (30 fps) rates. Can achieve
a compression ratio of up to 200:1.
DISADVANTAGES
Hardware-based playback limits distribution to suitably equipped
computers. Not a Video for Windows codec.
APPLICATIONS
Limited-distribution applications where playback hardwa
re is known
and sharp image quality is important (i.e., training, presentations,
and kiosks).
Truemotion S
From Horizons Technology, uses a scalable algorithm that runs under
Video for Windows. Software only. Capable of 640- by 480-pixel video
at 30 fps on Pentium-class systems, although performance on 486s is
much slower. Supports Windows, DOS, and Macintosh.
ADVANTAGES
Can run at resolutions and color depths higher or lower than the
original.
DISADVANTAGES
Performance is dependent on the host processor and bus type.
APPLICATIONS
Scalability makes it suitable for a range of applications, from
mass-market CD-ROM titles to high-end uses, such as sending
multimedia over LANs.
Video 1
Software-based, part of the Video for Windows codec library.
Originally developed by Media Vision, now owned by Microsoft. Uses
the proprietary Mo
tive technology for intraframe compression.
ADVANTAGES
Doesn't dither in low-color settings.
DISADVANTAGES
Lower quality than Indeo and Cinepak. Breaks up in high motion.
APPLICATIONS
Mass-market multimedia products and low-end machines. Works best for
low-motion video sequences and animations.
XingCD -- MPEG
XingCD is a software-based MPEG compression program. Image on right
was compressed with FutureTel's PrimeView II, a $17,000
hardware-assisted MPEG codec.
ADVANTAGES OF XINGCD
Performs MPEG compression with conventional processors, such as a 486
or Pentium. The $995 price is low compared to MPEG codecs using
dedicated processors (which cost from $2000 to over $50,000).
DISADVANTAGES OF XINGCD
Not Video for Windows compatible. Compression takes 10 to 100 times
the length of playback time.
APPLICATIONS FOR XINGCD
Multimedia CD-ROM developers.
N
ote:
Quality samples and test-result interpretations
were derived from the video compression sampler by Jan Ozer,
author of "Video Compression for Multimedia" (Academic
Press).
photo_link (153 Kbytes)
These photos show the differences in compression quality between the packages we looked at.
John Bryan is a technology writer and consultant based in San Jose, California. You can reach him on BIX or the Internet at
editors@bix.com
.
Compression Scorecard
