De-interlacing in Apple Final Cut Pro
Though many producers now exclusively shoot in progressive mode, few producers don't have some interlaced footage that they need to work with, whether legacy DV or some HD format shot more recently in interlaced mode. When producing for TV or DVD, interlaced works just fine. But when producing for streaming, or for film-based output, de-interlacing is necessary to convert the interlaced fields into progressive frames.
The problem is, between Apple Final Cut Pro, Compressor, and QuickTime conversion, there are at least five de-interlacing controls and or workflows that you can use to de-interlace your source footage, with potential combinations available. Here's what I'm sure is only a partial list.
- When working in a progressive preset, Final Cut Pro de-interlaces the footage automatically, though you can still apply Final Cut Pro's De-interlace filter.
What's the best approach? Well, it depends. For example, the answer is different if you'll be inserting the interlaced footage into progressive sequence in Final Cut Pro, or an interlaced sequence. It's also different if you're exporting with the Using QuickTime Conversion option rather than working through Compressor. While it's impossible to cover all the permutations, what I'll do this month is cover some of the most common workflow alternatives.
In this issue, I'll start with a quick overview of de-interlacing to help you identify and diagnose your problems. Then I'll look at scenarios encountered on the Final Cut Pro side of the equation, and when exporting Using QuickTime Conversion. Next issue I'll look at the Compressor side of things, primarily the quality and performance trade-offs associated with the de-interlacing controls found in Compressor's Frame Controls tab.
Let's begin at the beginning for those with no or only a hazy familiarity with the whole interlacing issue. In the beginning, all video was interlaced, which means presented in two fields displayed 1/60 of a second apart (1/50 of a second for PAL), rather than one complete progressive frame every 1/30 of a second (or 1/25 of a second in PAL-land). Each field contains alternative lines of the complete frame: the first lines 1, 3, 5, 7, and so on; the second 2, 4, 6, 8, and so on. Most importantly, these fields were actually shot 1/60 of a second apart, so they actually captured a slightly different scene.
Note that this was by design; the designers of NTSC television decided that it would look smoother if there were 60 distinct impressions rather than 30 (or 50 rather than 25). Nonetheless, this is the root of the problem irrespective of whether you want to use the fields to simulate film or to convert to a progressive streaming format. Specifically, when you convert the two fields shot 1/60 of a second apart into a single frame, they don't match. This is shown in Figure 1.
What you see in the figure are two fields, again, shot 1/60 of a second apart, combined into a single frame. However, in this high-motion shot, a lot happened in that 1/60 of a second, so again, the two fields don't match. Instead, you see the two most common problems that occur when interlaced video is not de-interlaced before converting to a progressive streaming format. First, you see the dual images of the skater's hand or the Hollywoodsports sign in the back. Second, you see the little slicing (or comb) artifacts that are evident all over. Either of these artifacts tell you that the two interlaced fields have been converted to a single progressive frame without de-interlacing.
So, what is de-interlacing? De-interlacing is the application of various algorithm and techniques to combine the two fields into a more integrated frame. As you can see in the figure below, de-interlacing algorithms can be quite effective, curing most of the visible ills seen in Figure 1. As evidenced by the Best designation in the upper left hand corner, this frame was produced in Compressor using the Best (Motion Compensated) option, which looks great, but takes roughly just short of forever to process.
Though you occasionally see some streams on the Internet that totally lack de-interlacing and look like Figure 1, most producers now know that they need to de-interlace. The problem is that de-interlacing techniques vary in terms of output quality and processing time. For example, as a preview of what you'll see in detail in the next segment, Figure 3 shows the same frame from three video files, processed using Compressor's Fast, Better, and Best de-interlacing options. If you click the figure to view the frame at full resolution, you'll see the quality difference between the three approaches.
In addition, as I mentioned at the top, Final Cut Pro offers multiple de-interlacing-related options. To get a feel for how these options compare in terms of quality and processing time, I used a file containing multiple sequences that either very quickly evidence when de-interlacing isn't applied (e.g. Figure 1) or contain content that's challenging to de-interlace and therefore highlight the qualitative differences between the alternatives (e.g. Figure 3). In short, jaggy-magnet scenes. You'll see various illustrative scenes from that file over the remainder of this and the next segment.
As you'll undoubtedly notice, most of the scenes contain some combination of motion and very crisp, clear diagonal edges, like the strings in the guitar in Figure 3. If your footage is dominated by low-motion content with few hard edges, like a talking head subject against a plain background, interlacing artifacts will be much less of an issue in your productions. You should definitely still de-interlace, but the qualitative differences between the techniques will be much less noticeable.
In addition, all of my comparisons involve a 720x480 4:3 DV file converted to a 640x480 streaming file, which is the most difficult test case. In contrast, when scaling footage downsay from 720x480 to 324x240, or 1080i to 640x360the qualitative differences between the different techniques are also much less evident because de-interlacing artifacts are reduced or eliminated by the horizontal scaling that occurs. For example, when scaling from 720x480 to 320x240, you can simply use one field or the other, so the de-interlacing issue becomes totally moot.
I should also say that I'm aware that there are many de-interlacing plug-ins for Final Cut Studio, some free, some relatively inexpensive, to help speed or improve your results. Perhaps we'll look at them down the road; this month it's Final Cut Studio-only capabilities.
With all this as background, let's get to scenario 1.
In this scenario, you're producing a video shot primarily with progressive footage, but are mixing in some interlaced footage. Obviously, you'll be using a progressive preset, which will de-interlace the footage automatically. The big question is whether or not you should apply Final Cut Pro's de-interlacing filter. The quick answer is that I wouldn't as a matter of course, since it's more of a blur-type filter than intelligent de-interlacing, but that it may improve quality in some instances.
According to Final Cut Pro's Help file, the De-interlace filter "Averages adjacent horizontal lines of the clip." That tends to reduce the jaggies, but also makes the image noticeably less crisp. Here are some examples.
In the chip cutting machine pictured in Figure 4, the De-interlace filter noticeably reduces the jaggies.
It also does a nice job reducing the wrinkles of our distinguished thespian in Figure 5, though clearly reducing the crispness of the entire frame. As an aside, I don't recommend shooting video in front of rattan curtains, but when you're shooting on location, sometimes you have to take what you get.
When deciding whether to de-interlace or not, processing time isn't a real factor: This 1-minute DV source test clip took 57 seconds to render to H.264 using QuickTime conversion without de-interlacing on my 8-core Mac Pro, while de-interlacing boosted processing time to 1:03. Rather, with Final Cut Pro's De-Interlacing filter, you have to balance the reduction in jaggies against the blurriness that the filter injects into the clip. If the jaggies aren't noticeable, I wouldn't use the filter.
In a situation where you're adding interlaced source to a progressive project, and jaggies abound, your highest-quality option would be to process the clip separately in an interlaced sequence and de-interlace in Compressor, though this adds time and complexity. You'll learn the workflow in our next segment.
I very seldom output from Final Cut Pro with the Using QuickTime Conversion export option, since Compressor is much more functional. When I tried this option on two different Macs, however, clicking the De-interlace Source Video checkbox in the Export Size Settings dialog did not work. I scanned around the Internet to see if I could confirm this observation, but found nothing on point.
If you need to export Using QuickTime Conversion from an interlaced Final Cut Pro setting, you'll need to either de-interlace using Final Cut Pro's De-interlace filter, or preferably export via Compressor, which will deliver much better quality. If you're working with a progressive setting in Final Cut Pro, check Scenario 1 to see if you should apply Final Cut Pro's De-interlace filter.
OK, that's it for this segment. Next time, we look to de-interlacing in Compressor.