Testing the New Windows Media Encoding Profiles
Entertainment
These clips include ballet, opera, country music, and a marimba player, with most clips having at least moderate camera and/or subject motion. Here again, Apple came in first (23 points total), with significant margins over both Flash (19 points, 17% difference) and WMV-AP (18 points, 22% difference). Still-frame quality was very similar with the top three codecs, with only one point separating them.
On the other hand, Apple proved four points better than Flash in temporal quality, and five points better than WMV-AP, though for different reasons. Again, WMV-AP performed very well on clips with low to moderate motion, especially when it came to minimizing noise in background walls. However, once motion became significant, blockiness got very severe. In contrast, Flash's downfall was background noise; fortunately, you can minimize this by choosing the appropriate background.
Quality Summary
Flash has made little progress since we last analyzed the format, dropping behind QuickTime in quality at these encoding parameters. Given the different value propositions offered by the two formats, it's hard to conclude that this is competitively meaningful. You can read more about this in Chris Hock's online interview on StreamingMedia.com. On the other hand, quality always matters, and this dynamic bears watching.
Obviously, Apple has also significantly improved its codec in the past year, improving competitive quality and eliminating random, highly distorted frames I experienced frequently when producing the codec survey in early 2006. Still, the bi-polar nature of its performance was puzzling.
Originally, I though that this disparity might have been caused by VBR encoding gone bad, though I had definitely opted for the "streaming" option in Compressor while setting my parameters. To be sure, I went back and verified that I had used CBR on the test file, and even encoded just the low-motion business file separately from the high-motion clips, and got very similar results. I also encoded the file with Sorenson Squeeze with similar results, though Squeeze doesn't have a 2-pass CBR option with the Apple codec.
Recognizing that Apple is widely used for movie trailers, I checked the average bits per pixel per second for movie trailers that I had viewed in the web survey referred to frequently in this article. I find the average bits per pixel per second metric a good way to compare videos compressed using different resolutions and data rates because as the name suggests, it measures how much data the producer allocated to each pixel in the video.
I found three movie trailers, for Borat, Babel, and A Prairie Home Companion, encoded into QuickTime format, and the average bits per pixel per second was 10.7. In comparison, the test employed here has an average bits-per-pixel-per-second of 3.3, about 30% of the bits used for the movie trailers. I'm sure if I tripled the data rate of these test files, the low-motion sequences in these files would look as good as the movie trailers. Overall, however, despite the Jekyll and Hyde performance, QuickTime/H.264 was clearly impressive.
So, clearly there's little reason to switch from Flash or QuickTime to WMV. On the other hand, the improvements in quality you can achieve by using the latest WMV codecs make switching to them a no-brainer. As long as you pay attention to the details, your audience will thank you.
Sidebar—How We Tested
Unless you're my mother you probably don't care how hard I worked to produce the ideal perfect test (and I have my doubts whether she cares). However, it does shed some light on several critical touch points in the streaming production workflow, and clarifies how the test is an objective measure of codec performance.
1. Friends don't Let Friends De-interlace in the Windows Media Encoder.
Our test involved converting a 720x480 interlaced DV file into a 640x480 progressive streaming file. This involves scaling and de-interlacing, the latter of which the Windows Media Encoder (WME) performs very poorly, as you can see in Figure 8. For this reason, you should scale and deinterlace your video first, then encode in WME.
Figure 8 (below). Windows Media Encoder is a competent encoder, but poor at deinterlacing.
2. When Quality Really Matters, Use a High-Quality Scaler/De-interlacer.
Premiere Pro, After Effects, and Final Cut Pro have integrated scaling and deinterlacing filters that provide good quality with most footage, but when your content has lots of diagonal lines and sharp edges, you can do better. On the left (see Figure 9) is footage processed by After Effects, on the right footage produced in After Effects using a plug-in called AlgoSuite from Algolith.
If you see jaggies in your footage after compression, often it's not the codec's fault, it's the program you used for the scaling/deinterlacing function.
Figure 9 (below). When quality really matters, consider a plug-in like Algolith’s AlgoSuite.
3. Choose a YUV-based intermediate format.
If you scale/deinterlace in a program other than your encoder, you need to choose a format for the intermediate file you'll create for inputting into the encoder. Codecs work in the YUV space, so it's best to choose a YUV-based codec like the Intel IYUV codec shown in Figure 10.
Figure 10 (below). Use a YUV-based codec like Intel IYUV for your intermediate files, not an RGB-based codec.
I've always defaulted to the Apple Animation codec, but Ben Waggoner says that a YUV-based codec is a better choice, because the Animation codec is RGB-based. In contrast, all codecs work in YUV space, and making the fewest number of RGB to YUV conversions helps preserve the color integrity of your video. I've never noticed a problem in the past, but I used YUV for the intermediate files created for this project, and will continue to do so for future streaming work.
I tried the 8-bit YUV (4:2:2 YUV) and 10-bit YUV (4:2:2 YUV) codecs also, but they wouldn't play back on the same computer that produced them, seemingly indicating that while the compressor component was installed, the de-compressor component was not. If experimenting with a YUV codec, make sure the files will load into your encoding program before finalizing your decision. This is especially important when you're encoding on a computer other than your production station, which I did here for my QuickTime encodes. More on that next.
4. Watch for Macintosh/Windows Production Issues
One particularly vexing issue was the black levels of our initial encodes on the Macintosh. Specifically, when encoding from the original source DV file, Final Cut Pro recognized similar black levels to Premiere Pro on the Windows platform. When I input the intermediate file from After Effects, Final Cut faded the image noticeably, irrespective of the format of the intermediate file (in addi