Digital Video Tests:

MPEG, MPEG2, Streaming, and AVI

E-Mail: Charles Branch
Web Page: Charles E Branch
We are looking at the tradeoffs in using digital video as an alternative to videodisc for Computer Assisted Instruction. We have tested MPEG1, MPEG2, AVI, and streaming formats using various compression/decompression CODECS and various hardware and software playback devices. This page contains some of our current experiments. For information on techniques, hardware, etc., see Summary and Conclusions.
 

The clips are in seven groups:

Introduction:  It is now possible, using affordable hardware and software,  to create high-quality digital video for Video-On-Demand applications or interactive Computer Assisted Instruction.  Soon, using the high-speed Internet2, it will be possible for schools to create applications which will be available to students at other schools--either on demand or in the form of live video conferences.  We will no longer be limited to the tiny jerky clips suitable only for simple animations or crude talking head applications.

DISCLAIMER:

These video files are large. They are not designed for online use in real lessons over the Internet. The quality of the MPEG clips will depend greatly on your hardware. For the best results, use a hardware MPEG decoder card, such as the SigmaDesigns Netstream2 or Hollywood cards we use. The clips are intended to illustrate some of the tradeoffs between quality, effort, file size, and hardware. When we get to the point of using live video over the Internet, we will need to look at some other alternatives.  NOTE: Some of the larger files may not be on the server at all times. If one you need is missing, please send me an E-Mail, and I'll put it up temporarily.

For our in-house video captures we use  DPS Perception card to capture video from either a videodisc, VHS videotape, or S-VHS videotape.  The Perception card captures the video in a 720x480 x 24 bit motion JPEG format.  The card can reproduce the video in what appears to other programs to be either a standard AVI file or a series of standard still images. From there we convert it to true AVI files or to MPEG using either Heuris MPEG Professional or CEQuadrat PixelShrink software encoders.  Recently we have begun working with the Canopus Amber real-time MPEG encoder card.

1. Compare Sample Clip Using Various InHouse Tests and External Vendors:  This sample is a short video clip which was originally created for our Cardiovascular Laboratory videodisc.  That page also contains some other sample MPEG video clips.
 
Capillary Perfusion Test:   This is an example of a Capillary Perfusion test.  The important point of this clip is to be able to observe the tissue blanch (turn pale) when you press on the gums, but return to normal color as the blood returns to the tissue. This is a simple test of the circulatory system. You be the judge of the tradeoffs between cost and quality. Can you see what you need to see in the video? Let us know what you think. 
These clips were captured from our  Cardiovascular Laboratory videodisc using the DPS Perception card and encoded using the Perception AVI file..

  • Hardware Encoding by External Vendors: These were generated from a Betacam SP dub of the original one inch videodisc master tape. The same tape was used by the three vendors. The products of only two of the vendors are shown here, because the third vendor's encoding was a very poor quality.
    • 2. Other InHouse tests using Heuris MPEG Power Professional Software Encoder
     
    High Resolution test using Polygraph Chart Recording
    MPEG2  Chart Recording.  3.7 MB.  InHouse (Mono Audio).  Note:  This clip contains only a few seconds from a 30 second clip.  The audio includes narration on the right channel and a cardiac beeper on the left channel
    Raptor Release
    MPEG2 Hawk.  2.0 MB.  InHouse. Converted from Videodisc using DPS PVR, DPS Editor, and Power Professional Version 1.3h in Windows NT. (No Audio).

    3. Blood Smear test:  Compare MPEG1 and AVI for file size and quality.
    These MPEG files were created using Heuris MPEGPro for the MPEG1 clips and VideoForWindows 1.1 for the AVI clips.  The two sizes are 352x240 and 160x120 for MPEG and 320x240 and 160x120 for AVI.  The MPEG files are smaller in file size, and probably higher in quality than the AVI versions.

    4. Hummingbird. InHouse MPEG1

    These MPEG files were created using the CeQuadrat PixelShrink program in either Windows 3.1 or Windows95.   PixelShrink is an inexpensive program which does high quality MPEG1 encoding, but has less flexibility than the more expensive Heuris MPEG Professional above.
    The original video was shot with an S-VHS Camcorder very early in the morning.  Some of the clips were shot using a shutter, which made the images darker than normal.  The videos were captured using the DPS Perception card.  The Perception "AVI" file was converted to either a true uncompressed AVI file or a sequence of TARGA images, and then encoded to MPEG.  Sample 1 was created from a normal S-VHS recording. The rest were produced from a different S-VHS recording which used a 1/100 speed shutter to stop the wings.

    Click here if you're more interested in this Hummingbird than you are in all this technical video stuff. Or Click here for even more neat Hummingbird information.

    5.  Hummingbird.  InHouse AVI

    True AVI Files encoded from the same captured Perception "AVI" file.  The Perception "AVI" file was first converted to a true uncompressed AVI file, and then compressed using various CODECs
    The original video clip was approximately 1100 frames captured from S-VHS at 720 x 480 x 24 bit x 30 frames per second. Using Video For Windows 1.1, an 88 frame clip was rescaled to 352x240 and then was saved as a standard uncompressed AVI file. That file was approximately 23 MegaBytes (appx. 8 MB/sec). All of the AVI test files were created from the 23 MB AVI file using Video For Windows 1.1. Another S-VHS clip shot without the shutter was also converted, but is not included here. However, the MPEG conversion of that file is included in the MPEG conversions below.

    6. InHouse Talking Head Test

    We are looking at the feasibility of using digital video for teaching communications skills for a colleague in the Department of Communications. Talking heads refer to images in which the background changes little from frame to frame. These types of images can be compressed more effectively than other types of video. Pans and zooms are more difficult, because the whole scene changes every frame. 
    An important consideration for talking heads is the requirement for accurate video and audio synchronization. One issue is the potential loss of good sync because the DPS card stores the audio and video in separate files which are merged later. That has turned out not to be a problem so far in our experience. 7. CVM Promotional Video:
    This is a 4 minute streaming video.  It was captured from a promotional videotape for the college's new Large Animal Teaching Hospital.  The MPEG1 version of this clip is 70 MB, and the MPEG2 version is 156 MB.  Those play immediately over our LAN using the Netstream2 card. The full video is included in the RealMedia streaming clip, and short 20-second snippets are also included as MPEG1 and MPEG2.
      300 kbps RealMedia:  8.75 MB (For access over LAN only)

      20 second Clips from Above Video

      A. Heuris MPEG Professional From DPS Perception "AVI" captured S-VHS Tape:

      B. Canopus Amber Captured Directly from S-VHS Tape.  The real-time capture is a great convenience, and our "out-of-the-box" experience was satisfactory.  We now use this card for most of our encoding, although we stiull use the Heuris program when time is not an issue.
    8. Video Conferencing: Most of us have seen or used video conferencing using the Internet and featuring small “postage stamp” video with perhaps several frames per second. Traditional Internet technology may be sufficient for some applications, such as talking heads, although even that is a problem during busy periods.  The commodity Internet is not sufficient for live transmission of subtle information, such as the abnormal gaits characteristic of certain neurological disorders.  The problem is twofold: First, the commodity Internet does not provide sufficient bandwidth for live high-quality video.  Second, the Internet does not even provide a guaranteed bandwidth at all, which results in frequent interruptions.  Until recently ISDN H.320 offered the only practical approach.  ISDN can provide a guaranteed specified bandwidth, but at considerable cost and complexity.  The advent of the Internet2 H.323 has made it possible to combine high quality with convenience and simplicity.  Video bandwidths as high as 1.5 megabits/second provide video and audio quality closely approaching traditional video, as well as the sharing of applications.  During the summer and fall of 1999 we conducted some experiments with colleagues at the University of Tennessee.  This was an extension of a project called "Virtual Rounds" proposed by Dr. Sims and his colleagues.  The experiments were a test of the feasibility live videoconferencing technology to conduct virtual rounds with simultaneous live participation of students, faculty, and patients at the two schools.
    Our conclusions were that Internet2 high bandwidth video conferencing is a viable technology for applications where god video quality is necessary.  Video conferencing using MPEG2 would offer even higher quality, but at substantially greater cost for both equipment and connectivity.

    Click HERE for some more information on our video conferencing project.

    Some test clips from our experiments are listed below.

    MPEG1  First Test
    MPEG2  First Test
    MPEG2  Second Test

    Summary and Conclusions

    MPEG Hardware Cards

    For MPEG2 playback we prefer the SigmaDesigns Netstream2 and the HollyWood or HollyWoodPlus cards.  The Netstream2 card is optimized for playing MPEG1 or MPEG2 files over a local area network from a server.  With the Netstream2 we  have played as many as six simultaneous MPEG2 videos from our server with no skipping or stuttering using only the MediaPlayer.  The SigmaDesigns Hollywood or HollyWoodPlus will also give good quality for both MPEG1 and MPEG2, but the files may tend to skip when played over a LAN.   In our experience the non-buffered cards tend to play video files better from a Novell server using IPX than from a comparable UNIX or NT server using TCP/IP.  We have also used the hardware-assisted MPEG2 player on the onboard ATI AGP graphics card in Dell 266 MHz Optiplex computers.  That system does an adequate job playing MPEG1 and MPEG2 video,  so long as the computer is not running other programs simultaneously.  We also had to trick the on-board sound card into thinking it was a real SoundBlaster.

    Our earlier work used several products which worked well for their time.  The Optibase PCMotion and Jazz Jakarta cards worked well for MPEG1 only. The Optibase PCMotion card outputs standard NTSC composite video. To display it on a computer we use a Compuvid SchoolBoard card to overlay the MPEG video onto the computer image; the SchoolBoard card has two software switchable inputs, which lets us compare the same clips in videodisc form and MPEG form on the same computer. Both the Optibase card and the Jazz Jakarta use standard 352x240 pixel MPEG video, but they can scale that image to fullscreen by interpolating, resulting in fairly high quality fullscreen video. The original SigmaDesigns RealMagic card plays MPEG1, but does not interpolate for fullscreen.   The newer Sigma HollyWood cards do a good job of interpolating to fullscreen.  We have not tested the SigmaDesigns Netstream1 card, which is designed to play MPEG1 over a LAN.


    Laptop Precautions:  One should NOT assume that computers which claim to have MPEG2 hardware-assisted playback capability can necessarily play standard MPEG 2 streams.  Having tired of lugging desktop computers to meetings for presentations, we purchased a Toshiba Tecra 750DVD Laptop for such presentations.  According to Toshiba sales persons, the 750DVD could play ordinary MPEG2 streams.  According to Toshiba promotional materials and the Toshiba manual, it has "special MPEG II hardware on the motherboard that can play full motion video files."  After many frustrating months seeking technical support from Toshiba, they finally informed us that the computer actually does not play ordinary MPEG II files.  We now use a Pro-Star laptop, which has a Sigma Designs hardware decoder card.  That card has since been discontinued.  We have also tested one software decoder program, but were not satisfied with its playback quality.

    We would be interested in the experience of other people who have laptops which are supposed to play DVD and/or MPEG2.  If you have such a laptop, please let me know your experience.  You could use the two test files test files above at http://www.vetmed.auburn.edu/~branch/dvtest/#compare.  Or you could also get them at ftp://ftp.vetmed.auburn.edu/pub/branch/MPEGtest/.   It would help a lot if you would answer the questions below.  Just cut the questions below and paste them into an E-Mail to me at branch@auburn.edu
    Maybe we will convert this to a form, if there is any interest.

      1. Computer type (laptop, desktop, lunchbox) _________
      2. Computer Brand/Model ______________________________
      3. CPU frequency if known ________________________
      4. Memory if known __________________
      5. This computer is supposed to play DVD and/or MPEG2.  _____ yes   _____ no  _____ don't know.
      6. MPEG2 Playback Hardware if known ________________________
      7. Does it play using the Media Player (MPlayer.exe)?
          MPEG1 ______
          MPEG2 ______
      8. Does it play using ActiveMovie (MPlayer2.exe)?
          MPEG1 ______
          MPEG2 ______
      9. Did both audio channels play? ______
    10. Was the video quality adequate to observe the perfusion? ______
    11. Your E-Mail address (optional) _________________________
    12. Other comments:

    MPEG File Creation Procedures

    Video Capture:  Most of the InHouse clips were originally captured using a DPS Perception Video Recorder. The DPS "AVI" file is actually a pointer table which references the DPS proprietary motion JPEG file.  Programs such as VideoForWindows and Adobe Premier can use the file as if it were a true AVI file.  The Heuris encoders also can use the DPS file; that eliminates the laborious step of creating a true AVI file, and also eliminates having to temporarily store a very large uncompressed file.

    Heuris MPEG Professional:  MPEG Professional is a high-end software encoder which can produce MPEG1 and MPEG2 files comparable in quality to those we have had produced by commercial vendors.  The program can use the DPS "AVI" files, making it possible to convert from the DPS file directly.  MPEGProfessional has many options for controlling the encoding, such as control of bit rate and inserting I-frames at will.  It can even use two-pass analysis to adjust the encoding based on future frames.

    Canopus Amber for direct real-time MPEG1 and MPEG2 encoding.  We have been quite satisfied with this card.  The real-time encoding is a very great advantage.  This would make it possible, for instance to record classroom lectures and make them available for review over a campus network on the same day.  The Amber software has fewer options than the Heuris software encoder.  Since it is a real-time single-pass system, it can not analyze the video stream and make encoding optimizations based on future video frames.

    CeQuadrat PixelShrink: Some of our early encoding tests used PixelShrink, which was a lot cost MPEG1 encoder that could convert AVI files to MPEG1.  PixelShrink could not convert the DPS "AVI" files directly. The encoding steps consist of 1) Capturing the live video, 2) creating a PVR "AVI" file, 3) Converting to an uncompressed standard AVI file, and 4) Converting the AVI to MPEG.

    1. Record the live video (S-VHS) into the DPS Perception card at 720x480x30 frames per second, . This part works great. The video is virtually the same as the analog videotape.
    2. Use the DPS software to Convert to a Perception "AVI" file at 720x480x30 frames per second, which is really a sound file with pointers to the Perception video file. This is also fast and simple, since it is not having to do any actual AVI CODEC compression.
    3. Use VideoForWindows 1.1 to rescale to 352x240 and save as an uncompressed true AVI file. NOTE: This is a really big file, and the conversion is slow. It would be much nicer if I could trick the MPEG encoder to deal directly with the Perception "AVI" file.
    4. CeQuadrat PixelShrink: MPEG Conversion: This is a little confusing. CeQuadrat recommends Windows 3.1 rather than Win95. I tried both, and both seem to work. Win95 appears to produce a correct file with the correct 352x240 MPEG screen. Win31 also works, but the file winds up with a different aspect ratio, which gives a black space at the top (kind of like you see on some movies on your VCR). The file also seems to be larger.

    Overall Conclusions:

    1. The PVR card produces extremely high quality (comparable to S-VHS video).  The DPS method of accessing "AVI" files eliminates the 2GB limit in file size for ordinary AVI files.  The files can be edited in real-time, so long as you do not use dissolves or other special effects.
    2. The PixelShrink software encoder produces quite acceptable MPEG video for our application. It would be MUCH better if it could skip the step of creating the extremely large uncompressed AVI file. The clips do behave strangely at the end. Some of the other test clips using external vendors behaved better. Some of the hardware encoders were able to insert key frames as the last frame, which made the clip stop at exactly the desired location.
    3. Heuris Pulitzer's Power Premiere and Power Professional software encoder also produce quite acceptable MPEG1 video for our application. MPEG2 encoding is an option with MPEGPro.  Using a dual 350MHz  PentiumII, the MPEG1 encoding speed is about 16.7 times real time, and the MPEG2 encoding is about 32 times real time for video, audio, plus multiplexing.  The jobs may be batched for encoding overnight. The Heuris programs have the advantage of bypassing the step of creating an uncompressed true AVI file, and thus eliminates the 2GB files size for AVI files. It also is more versatile than PixelShrink. For instance, we could insert an I frame at the end of the short clip, which appeared to make the clip end more predictably. This program, however is many times more expensive than PixelShrink, and it requires the use of a hardware key (dongle) on the parallel port.  This encoder produces software MPEG2 encoding comparable in quality to that we have observed on commercial DVD recordings.
    4. Real-Time MPEG encoders would reduce the time required for encoding, and would make live streaming MPEG feasible.  The cost of MPEG encoding is coming down rapidly.  Economical real-time MPEG2 encoding using the Canopus Amber and other cards is now available.

    Last Update: June 12, 1999