This article 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. Hidden in the suspenseful plotline are some invaluable lessons about shooting for streaming, ways in which different timeline presets can impact compressed quality in some programs, and which format and settings to use when exporting intermediate files from your editor for encoding elsewhere. Who says learning has to be dull?
The story begins in the form of a running debate between myself and a colleague named Tim Siglin, an industry analyst and consultant. Tim claimed that shooting in HD for streaming delivery produces dramatically better results than shooting in SD, even if the target resolution of the streaming file is smaller than SD.
Most HDV camcorders can also shoot in DV mode. To refresh your memory, HDV has a pixel resolution of 1440×1080 with a display aspect ratio of 1:33, which means that the horizontal pixels are zoomed out to 1920 for a full-resolution 16:9 display of 1920×1080. In 16:9 mode, DV is captured at 720×480 pixels with an aspect ratio of 1:2, so the horizontal pixels are zoomed out to 864 pixels for a full-resolution widescreen display of 864×480.
Assume that you were planning to stream a 16:9 file at 640×360 resolution. Tim’s claim was that you’d get significantly better results by shooting in HDV. I claimed the results would be indistinguishable from DV, and my standard practice is to shoot in DV mode when shooting for streaming in less-than-SD resolutions.
This debate has gone on for some time, but I’ve never run comparative tests because I’ve never had two identical camcorders at the same time to run them. Obviously, with two different camcorders, any qualitative differences could relate to the camcorders, not the difference between acquiring in HD vs. SD.
I could have run the test sequentially with the same camera, shooting first in one mode, then the other, but felt the test needed a live component which you can never duplicate when shooting sequentially. Long story short, when fellow shooter Carrie Cannaday called to say that she was coming by to visit the Galax, Va., Old Fiddlers Convention this past week, I asked her to bring her Canon XH A1—which gave me a matching set.
The high-level test setup was this. I created a simple scene in my office, just me and the DSC Labs ChromaDuMonde chart. Then I shot the scene with two different camcorders—one in 16:9 DV mode, the other in 1080i HDV mode. Both were in Canon’s Frame mode, which is the company’s version of progressive, and all other camera settings were identical.
Figure 1. My initial results from Adobe Premiere Pro.
Click here for a larger image
After the shoot, I created separate projects for each source video in Adobe Premiere Pro, one at DV resolution, one at HDV resolution. That way, I could capture and process each file at its full resolution, which I then did. Then, I exported an H.264 file from each Premiere Pro project at 640×360 resolution, 30fps, at a data rate of 600kbps—which should be more than sufficient for the talking-head video.
Then I compared the two files, which you can see in Figure 1, and I found that the video from the high-definition source had noticeably better still frame quality and fewer artifacts during playback. If you look at Figure 1, you’ll notice that the text in the HDV frame is noticeably sharper, that the picture frame to my right has much less aliasing, and even that my face seems to “pop” much more noticeably on the HDV source video.
Figure 2. Results from Apple Final Cut Pro and Compressor.
Click here for a larger image
In short, things were looking grim for our hero, and I starting thinking about recipes for the crow it looked like I was going to be eating. But something didn’t feel right about the results. So I duplicated the tests in Apple Final Cut Pro, and I found the resultant frames and video almost identical in quality, as you can see in Figure 2.
Gone was the aliasing around the picture frame and the noticeable disparity in the detail in the chart. Gone were the motion artifacts during realtime playback. Gone was HDV’s superior clarity around the face.
The obvious questions started to come, fast and furious. Was I doing something wrong in Premiere Pro, or could I somehow duplicate Compressor’s quality? What would happen if I encoded the files in a third-party program such as Sorenson Media Squeeze, and what settings would I use to create the intermediate file in Premiere Pro? Was shooting in HDV best practice or not? Most importantly, would I have to admit defeat when I next met with Mr. Siglin?