For image quality buffs, DSLR video is off the table

The 5DmkII is, it should be said first, an excellent camera, and it takes amazing still photos. Many people seem to think, though, that along with their professional still camera they are getting a professional video camera. While the video it takes is leagues ahead of cheapo handheld HD cams like the MinoHD and Zi6, the video is fundamentally lower-quality than true, dedicated digital cinema cameras. This wouldn’t need to be said if people weren’t saying otherwise, and it’s no slur against Canon either, but if misconceptions like this are allowed to continue, that’s a problem.

Above you see a crop of a zone plate, a set of concentric circles often seen in photographic testing (like here but imagine circles) to determine sharpness, chromatic aberration, and that sort of thing. The image above is a still from some 1080p 5DmkII video taken by Jim Jannard, creator of the RED camera which the 5D is mistakenly compared to (partially RED’s fault). Here’s a resized version of the full image:

As you can see, there is some really egregious Moiré effect going on. Good example of how the circles should look here. Note that Moiré is seen in almost any capture of a zone plate or focus test, but this is particularly bad. What causes this? An assessment by Pango at the forum based on the sensor size and resolution suggests that the 5DmkII keeps every third line in the 16:9 AR area and resizes it horizontally to 1920 pixels, creating a 1920×1080 image. No one knows for sure, but something like that is almost certainly going on. The following image is very much an exaggeration, since normally you’d have much more information to interpolate, but it gets the idea across:

This method of throwing away lines, resizing and recombining them is of course much more noticeable when you can check out the individual frames and when the pixels are made super-visible, as above. However, when the technique is used to resize video, it results in a predictable pattern of aliasing, or jaggy lines and unnatural motion, along the edges of things that are diagonal, most so when they are close to horizontal. Near-vertical lines, faces and some details are interpreted all right, but edges signs, horizon lines, building edges, all these things, depending on their orientation, may have a shimmer or stepped quality to them. The Moiré pattern seen in the concentric circles above is an indicator of this problem, and shows at what angles and with which colors is it most likely to happen.

It’s true that once resampled down to plain ol’ HD or lower than that, it’s nearly imposible to notice, and perhaps it’s not so easy to see in the first place if you’re not looking for it. But for real cinema, it’s an insuperable barrier for DSLR video right now because they can’t output video from their whole frame — thus the need to use every third line, or whatever technique is in use. I don’t have a Moiré detection plate for a Nikon D90 or I’d have mentioned it more prominently; it may be that it’s better, or worse, although the sensor of the D90 appears to read slower, resulting in more skew and jelly-motion.

I’m not sure how many people were really making a big deal about the new set of DSLRs being professional video cameras, but the few movies we’ve seen come out sure aren’t doing anything to dispel that notion (other than the first one being really bad). If you were to see 5D or D90 video at full size projected in a theater, the disparities would be obvious.

I just thought I’d throw this out there so people know what’s going on that makes a professional digital cinema camera different from a professional still camera that shoots video. Doubtless the problems go both ways as well: the RED may show up the 5DmkII in video, as well it should costing over ten times as much, but it can’t touch it for still image quality and capability. Let’s just stop comparing the two altogether. DSLRs will get better and their video is great for basic stuff, but there’s just no comparison.