Recently Sorenson released the Sorenson Video 3 CODEC. According to Sorenson, this CODEC is a vast improvement over Sorenson Video 2, giving higher quality video at the same bit rate with faster compression and making better use of VBR (variable bit rate). Currently, Sorenson Video 3 is available in two forms; Sorenson Video 3 Standard Edition, which is included in the latest release of QuickTime 5.0.2, and Sorenson Video 3 Professional Edition.

The Professional Edition has many more features over the Standard Edition, including:
· Improved performance
· Enhanced quality
· Support for alpha channel/chroma key
· Color watermarks
· Automatic keyframes through scene change detection
· Bi-directional prediction
· Support for two-pass Variable Bit Rate (VBR) compression
· Block refresh for packet loss correction
· Media key support through secure encryption
· Compression time packetization for error resiliency to packet loss

Of course reading about Sorenson Video 3 gives you one impression. How does the new CODEC actually work?

I haven't received the Sorenson Video 3 Professional Edition as of this writing, but I was eager to check out the compression quality available in the Standard Edition. If you follow DMN TV, our streaming media show, which is part of the Digital Media Net family, you may have noticed our latest edition is encoded with Sorenson Video 3. Co-Producer Paulo de Andrade and I were both amazed at the quality we were able to achieve with this new CODEC and have decided to encode future editions of DMN TV in this same format.

But how much better is SV3?

I took some time this week to encode some footage supplied by our friends at Art Beats from their Digital Film Library to find out. I selected three clips for this test, each for their unique qualities that can make compression and streaming difficult.

From the Reel Fire 2 collection, I used the brushfire ignition clip (RF202). This clip offers many hurdles for encoding, including fast motion, detail and huge areas of contrast, all of which which can make video compression a nightmare.

Because not everyone needs to compress video that has these specifications, I compressed a cloud fly through from the Cloud Fly-Thrus 2 collection (CF217). While this video clip does have some drastic motion, that is typical in many streams today, it also has subtle color and tonal variations that can be lost when a video is compressed for the Web.

For the final clip of this test, I wanted something that combined both of these, so I chose a canyon fly over from the Aerial Landscapes collection (AL110A).

Incidentally, if you would like to view each of these in their pre-Digital WebCast forms, visit www.artbeats.com and use the reference numbers above to access the clips.

The Test
Because I only have access to Sorenson Video 3 Standard, the only compression option available is the Basic compression option. I previously have been using the Sorenson Video 2 Professional Edition, and the two have very different variables, which can be set in Cleaner 5. To try and make the tests come out as close to one another as possible, I used the secret DMN TV compression settings that were used in the latest edition of DMN TV (www.dmntv.com), which happens to be compressed at 15 FPS at 240 x 180.

I also want to make it clear that these tests are being done with the Standard version of Sorenson Video 3. Results should be quite a bit different when using the Professional Edition.

I compressed these files three times; Sorenson Video 3 Basic, Sorenson Video 2 Professional Edition with Basic settings and Sorenson Video 2 in VBR.

The first round of tests were done at 300 kbps which is a speed designed for viewers on a DSL or Cable modem. At this rate, all of the video looks pretty good, with only slight variations in the canyon and clouds example. The biggest differences can be seen in the fire example. As far as over all quality goes, Sorenson Video 2 Basic comes out on top in this example. While it is slightly "fuzzier" than the other two, it maintains the frame rate and contrast over the others. Sorenson Video 2 VBR doesn't really come out well in this series, and while Sorenson Video 3 Basic maintains the best detail; for some reason the compression falls apart in the last moments of the video. (I reran the compression several times with the same results.) I should also point out that no keyframes were set in these examples, which would have improved the quality of each of these clips.

The big difference that can be seen in these examples is in file size:

300 kbs file size results
Canyon
Sorenson Video 2 Basic 902KB
Sorenson Video 2 VBR 871KB
Sorenson Video 3 Basic 866KB

Clouds
Sorenson Video 2 Basic 536KB
Sorenson Video 2 VBR 513KB
Sorenson Video 3 Basic 498KB

Fire
Sorenson Video 2 Basic 472KB
Sorenson Video 2 VBR 373KB
Sorenson Video 3 Basic 566KB

Because of the fast motion, contrast and other aspects of the fire video, Sorenson Video 3 Basic does fall behind the Professional settings of Sorenson Video 2. However, Sorenson Video 3 Basic does beat the other two in the other examples. While this may not seem like big differences, remember these are short video clips. Over the long run the file size savings will add up.

While it would be nice to believe that everyone in the world is happily sitting at home with a fast connection, that really isn't the case as many people are still on 56k dialup. Because of this, I wanted to see how these same clips would look at a lower compression setting. In this next set of compression tests, I left all of the settings the same, but lowered the rate to 100kbs (56k dialup). This is where Sorenson Video 3 Basic shines.

Again, in the canyon and clouds test, Sorenson Video 2 Basic and Sorenson Video 3 Basic are pretty close. Sorenson Video 3 Basic is slightly softer in the canyon clip than 2, but artifacts and smearing show up in the cloud sample when compressed with Sorenson Video 2 Basic, and not in the Sorenson Video 3 Basic CODEC. When examining the samples compressed with Sorenson Video 2 VBR, many artifacts, color deviations and softening of the videos make these settings fall behind.

The best example of how much better Sorenson Video 3 Basic is over the other two settings can be seen in the fire example. Sorenson Video 2 Basic falls apart very quickly, and Sorenson Video 2 VBR compressed the sample into pixelation hell.

100 kbs file size results
Canyon
Sorenson Video 2 Basic 358KB
Sorenson Video 2 VBR 291KB
Sorenson Video 3 Basic 290KB

Clouds
Sorenson Video 2 Basic 247KB
Sorenson Video 2 VBR 171KB
Sorenson Video 3 Basic 168KB

Fire
Sorenson Video 2 Basic 248KB
Sorenson Video 2 VBR 124KB
Sorenson Video 3 Basic 171KB

Final Thoughts
If you are someone who is working in streaming media, then you owe it to yourself to do some Sorenson Video 3 Standard tests yourself. I think you will be just as pleased as we were when we began compressing video and audio for streaming at Digital Media Net. The best thing for those working in streaming media with limited budgets is that Sorenson Video 3 Standard is free. You can't beat the improved quality for cost there.

If the results we have gotten with Sorenson Video 3 Standard are any indication, the Sorenson Video 3 Professional CODEC should produce some excellent results in VBR and other compression setting modes. When I review Sorenson Video 3 Professional, I'll run these tests again and post the results in the review. For more information on the Sorenson 3 Video CODEC, visit Sorenson.com