Shooting for Streaming, Part 2

In our last episode, we faced a paradox. Producing a streaming file of less-than-SD resolution using format-native presets in Adobe Premiere Pro showed much better results for HDV source files than for DV. Yet duplicating the test in Apple Final Cut Pro revealed nearly identical results for HDV and DV. Was this a flaw in Premiere Pro or an error on my part? Does HD or SD produce better quality video when streaming at less-than-SD resolutions?

My first thought was that I had used the wrong DV preset in Premiere, so I duplicated these tests. I got the same result. Then I wondered if the results would be any different if, rather than use the format-native presets for the projects, I used a 640x360 preset in Premiere Pro. This is my standard practice when producing for streaming, although more for convenience than any quality-related issue. Let me explain.

When producing for streaming, you have two logical choices for the project preset. You can use a preset that corresponds to the format of your source video or use a preset that matches your target output—in this case, 640x360. I like the latter approach because it provides the best possible preview of your finished content, which is important when creating and placing titles and even for framing your videos within the target window.

Figure 1. Creating a custom preset in Adobe Premiere Pro.
Click here for a larger image

You create a custom preset when starting a new project as shown in Figure 1. Obviously, you insert the target frame size in the appropriate fields, but you should always make sure to choose the Square Pixels (1.0) pixel aspect ratio, even if your source video is DV or HDV and has a different pixel aspect ratio. You should always choose No Fields (Progressive Scan) as well, because most streaming files are progressive.

Figure 2. Resizing the source video to fit in the smaller project.

Click here for a larger image

Once you create your preset and start to edit, note that you''ll have to scale the video to a lower resolution to make it fit in the smaller window. You can see this in Figure 2, where I scaled the original DV video to 75 percent of the original size. Or you can leave the video at 100 percent and reframe the video within the window—say, if you hosed rule-of-thirds positioning or you want to convert a medium shot into a closeup. As long as you don''t zoom higher than 100 percent, the video will be perfectly clear.

Figure 3. What a difference a change in preset makes.
Click here for a larger image

Long story short, after creating the second project and getting my video situated, I re-rendered the file. I found the results much improved, as shown in Figure 3. This is most obvious in the picture frame, which shows much less aliasing. Although it's more subtle, if you click to see the original image at full resolution, you''ll notice additional detail in the chart—particularly in the horizontal patterns on the corners of the colored grid.

Related Links

Shooting for Streaming, Part 1 This month''s offering is a drama showing just how far a man will go to prove that he's right (or perhaps, not wrong). What''s in it for you besides the guilty pleasure of reading a desperate nail-biter? It''s called stealth learning...

Why would the preset make a difference? I have no idea, and tight editorial deadlines and a looming vacation make it highly likely that I won''t find out before you read this article. Still, it tends to add credence to my suggestion that streaming producers working in Premiere Pro use a preset that matches their target output. Note that I went back and tested Final Cut Pro with a 640x480 preset, and I produced a file identical to the original file—which I produced using a DV preset.

Before finalizing my conclusion about whether it was better to shoot in HD or SD for sub-SD resolution streaming, I wanted to encode the DV and HDV footage in a third-party encoder and compare the results. Because Sorenson Media Squeeze is a relatively inexpensive cross-platform solution, it seemed like the best alternative.

However, I had edited both source videos in Premiere Pro, trimming, brightening a bit and (sigh) color-correcting the source. So now I needed to export an intermediate file from Premiere Pro to import into Squeeze. In general, when producing an intermediate file from an editor, you have two high-level choices.

First, you can render a file at the resolution of the original acquisition format, which forces the third-party encoder to scale and often de-interlace the video to its final resolution when producing the streaming file. Alternatively, you can render the intermediate file at your target resolution, letting the NLE scale and de-interlace, and simply encode in the third-party encoder. Normally, I use the latter alternative, but because scaling quality seemed to be the crux of my comparison, I used the former method to force Squeeze to scale and output both files. Within Premiere, I kept the files at their original native resolutions (using the DV preset for this operation for the SD video, rather than the 640x360 preset).

With either alternative, you have to choose a format for storing the video. One alternative is to simply store the file in its original acquisition format (HDV if HDV, DV if DV)—which is simple, but not best practice because the NLE will have to re-compress the video if you''ve made any global change such as the brightening and color correction that I had performed. Better to use a lossless or less lossy format.

When choosing an intermediate format, you have several considerations. First is the length of the video file itself, which can impose some practical limitations. If you''re working with a full-length movie, using a lossless format such as Apple''s PNG codec or the Lagarith YUV12 codec may not be workable; you may need a lossy format such as Apple''s ProRes 422.

Assuming your video is relatively short, like this one, you can use a codec such as the QuickTime PNG codec, which is great for cross-platform encoding (which I do a lot). Or for Windows-only work, the Lagarith codec is a great choice.

I used to use the Apple Animation codec to produce files for cross-platform encoding until Microsoft''s Ben Waggoner pointed out that this codec was RGB-based, while most origination formats and streaming codecs are based in YUV color space. He (and several other sources that I''ve since read) recommend against this codec because the additional YUV-to-RGB conversion can cause distortion, although I''ve never seen this in practice. Still, at Ben''s suggestion, I''ve starting using the QuickTime PNG codec to produce files compatible on both platforms, and it''s about 50 percent more efficient than the Animation codec with equal quality.

Figure 4. Encoding to the same parameters as HDV but using the PNG codec.
Click here for a larger image

When exporting the intermediate file, you have to be careful to match the original specs of the acquisition format except for the codec, and Figure 4 shows most relevant details of encoding the editing HDV file into an intermediate file for input to Squeeze. Note the frame size of 1440x1080 and anamorphic pixel aspect ratio of 1.333:1, which matches HDV''s attributes. Here, the original file was progressive, but if it were interlaced, I would have chosen that in the Keyframe and Rendering tab.

Note that if you were producing an intermediate file to match your target, you would use parameters that closely match Figure 1. That is, you''d input your target resolution and specify square pixels, not the aspect ratio of your source video.

As a rule, I also produce all intermediate files using Premiere Pro''s File > Export > Movie sequence, rather than the Adobe Media Encoder. Although Adobe tells me that the results should be identical, the Export Movie sequence is easier to follow and feels more reliable. I have gotten some funky results when producing intermediate files with the Adobe Media Encoder—particularly with screencam source videos, where any distortion is immediately obvious.

Figure 5. Sorenson Squeeze''s DV and HDV results are very, very close.
Click here for a larger image

OK, so I created the intermediate files as specified and encoded in Squeeze, and Figure 5 shows the results, which were the closest yet in terms of both detail and motion quality. I like the HDV color a bit better, but the faded-blue background in the DV clip is more accurate. Both are easily configurable via Premiere Pro''s (or Final Cut Pro''s) color filters.

So, what''s my conclusion? All things being equal, no viewer could detect the difference between video shot in SD or HDV if shot with the same camcorder. With side-by-side comparative displays, HDV video processed in Premiere Pro looks just a hair better than DV—although video produced by Squeeze and by Final Cut Pro look nearly identical.

Note that this result might change if you were comparing DV with DVCPRO HD or AVC-Intra, which offer higher-quality HD than HDV. But for HDV vs. DV in the same camcorder, DV should produce nearly identical quality as HDV.

Getting back to the original debate that started this process, we get the best possible result. Tim could be right, and perhaps will be proven so if I ever get my hands two really high-quality HD camcorders. As for now, at least as it relates to DV and HDV, I am right, although obviously only if you use the optimal presets and procedures all the way through.