Buying a Canon DSLR. 60D, 7D, or 5D Mark II?

[Gary Huff]

You are right it’s not simply from AVCHD. The AF-100’s sensor is primarily designed for video (unlike HDSLR) so there is no moire or aliasing like the Canons, plus a bit better on the resolution side.

[Dave Haynie]

[Gary Huff] “You are right it’s not simply from AVCHD. The AF-100’s sensor is primarily designed for video (unlike HDSLR) so there is no moire or aliasing like the Canons, plus a bit better on the resolution side”

I wouldn’t read too much into the “designed for video” idea. Yeah, the camera as a whole, like any video camera, is nicely adapted for video, versus any HDSLR, which is optimized for SLR use. Things like the variable frame rate help push it into real pro territory. Keep in mind, the video feature in the 5DMkII was originally kind of a “because we can” thing… they didn’t even bother doing 24p until, based on the huge response to the camera, they dropped that into an upgrade. DSLRs haven’t exactly remained frozen since the initial release of the 5DMkII. Though Canon’s still locked into the stupid 12 minute/4GB limit, even on exFAT cards. This is the primary reason I could never use an HDSLR as my primary video camera. But hey, if you’re used to 35mm film reels…

The sensor, though… same sensor used in the Panasonic GH1, I believe… it’s a 12Mpixel 1/2 frame Four-Thirds sensor (Micro-FourThirds configuration, but that’s only the lens mount… sensors on four thirds cameras are all the same size). If you’re after that cinematic depth of field, this won’t be quite as good as the Canons. The new Sony NEX-FS100, on the other hand, slightly larger sensor than the Canons. With that said, I’ll admit all my current camcorders are Pannys, and my SLRs are Canons, going back to film. So I’m unlikely to care. Much.

The big win in the AF100, in theory anyway, is that, in front of that sensor, it sports an optical anti-aliasing filter, which you don’t have in an HDSLR. That helps to reduce resolution, sure, but also aliasing noise… which, worse-case, is the moire pattern noise that’s sometimes visible in HDSLR video. The HDSLR is designed to shoot stills at full resolution, so it can’t have the anti-aliasing filter, unless you swapped one in… you’re going to see the aliasing, at the very least, when the original 18Mpixel is resized to 2Mpixel. Most normal smaller-sensor pro video cameras don’t have this issue, since they’ll just have sensors that exactly match the video resolution.

The other issue is that, even with the filter, there’s still money business going on. You want at least 8Mpixel of sensor detail to allow for pixel bucketing rather than Bayer interpolation. Given the 12Mpixel and original aspect ratio, I bet it’s just about 8Mpixel when masked to 16:9. That color interpolation, while not as bad as going from the 18-21Mpixel sensors on the Canon, will lead to some anti-aliasing evil you wouldn’t have on a normal 2Mpixel 3-chipper.

They don’t do anything to deal with the “jello” effect, other than the differences between cameras in general (faster CMOS processing of the image, less time for movement, less jello). The problem isn’t specifically CCD vs. CMOS, but the fact that CMOS sensors built so far use a rolling electronic shutter, rather than a global shutter. A CMOS sensor designed specifically for video could support a global shutter. CCDs do, largely as a means of reducing other problems — they effectively buffer the whole image. The problem with that is that there are actually twice as many charge coupling devices as you’d need for the sensor alone, so it gets expensive. And it would, for a CMOS sensor too. Another option would be to use a mechanical rather than electronic shutter, though CMOS sensors pretty need an electronic shutter anyway to do noise cancellation (the entire sensor gets a zero light exposure, which eliminates all patterned noise inherent in the sensor).

AVCHD is not a higher-end version of H.264, not even slightly. Like most MPEG specs, H.264 (aka AVC, aka MPEG-4 Part 10) is simply a compression scheme, not a whole spec. AVCHD was defined specifically for consumer cameras, to fully specify the rest of the format, for easy editing support. AVCHD 1.x is limited to 24Mb/s encoding, 1080/60i, 1080/24p, 720/60p, 720/24p… curiously, 1080/30p isn’t actually supported… if you have that in your camera, you’re actually getting 1080/60i encoding. The AVCHD spec was drawn as a proper subset of the Blu-ray spec. Since it was originally for recording on DVDs (as I said, it was for consumers), the original version allowed only 18Mb/s video, and that’s still the limit on DVD. The other characteristics — you get a whole BDMV style file system as part of the spec, though it’s technically not exactly BDMV, there are a few trivial differences. The video of course is stored in an MPEG-2 transport stream, and audio must be encoded in AC-3, video in H.264 (Blu-ray, obviously, allows other video and audio options).

Most HDSLRs encode H.264 in a Quicktime wrapper, which is also very standardized, but not limited in the rates supported by AVCHD. There’s nothing intrinsically different about the H.264 that’s encoded… it depends on the camera. Many AVCHD camcorders only encode at H.264 Main Profile, some do encode at High Profile, which can be ever so slightly more efficient. Another constraint of AVCHD… it had to work with Class 4 SDHC cards… even the new version, AVCHD 2.0, which officially supports Panasonic’s 28Mb/s 1080/60p mode (and some goofy 3D stuff too) still maintains that Class 4 support.

There are a few quirks in dealing with Canon, anyhow, while I think Panasonic, in HDSLRs and Camcorders, pretty much goes by the book. For example, Canon is using full swing YUV in their H.264 encoding, but many if not most decoders assume that any YUV encoding is automatically studio swing. What you’re looking for in the Canon’s (and some other HDSLRs, including Panasonic’s) is a limit in the types of Intraframes the encoder will create. H.264 supports three kinds of frames: I (independent), P (predictive), and B (bidirectional). If you’ve seen camcorders that record in AVC-Intra, they use I-Frames only, usually recording 4:2:0 at 50Mb/s or 4:2:2 at 100Mb/s…. it’s kind of a glorified version of DV or Motion JPEG… each frame is separately encoded.

Predictive frames encode a set of motion vectors that can be applied to the previous frame to mostly encode that P frame. They then store just the differences, which is usually very small. B frames work the same way, but can be based on either prior or next frame, so they can be a little smaller still. The Canons all use only P and B frames. Some HDSLRs did this at 17Mb/s (some of the early Pannys, I think) and yeah, poor quality video. But the 44Mb/s (average) of the Canon, this produces better quality video than 50Mb/s AVC Intra. You’re not going to better that with IPB encoding at 24Mb/s. Thing is, both of these are superior to HDV and other forms most of us are used to, so we tend to be pretty happy with either one. Though the Canon HDSLR video is a bit computationally easier to edit native than full IPB video.

There’s also no requirement that AVCHD encoders produce IPB video, either… that’s entirely up to the implementation. IPB, IP, and I-only video are all legal H.264 video streams. It’s also the case that H.264 encoders have been improving yearly. I’ve been playing with H.264 cameras for six years or so, and until about 2009, they didn’t compete with MPEG-2 cameras. It’s not the spec, it’s the ability of the on-camera signal processor to do this encoding for about 3W of power. It’s the manufacturer’s experience with the algorithm. It’s even the video you’re shooting… AVC and MPEG suffer with fast motion, but higher bitrate cameras may deal better with high detail, low motion scenes that stymie a lower bitrate encoder. And it’s still pretty true that this year’s model does a better job than last year’s, though it’s settling down.

I think the AF100 is a fine model, but it’s not the last word in video. And I think there was a great deal of disappointment that Panasonic didn’t go further… like, why not an AVC Intra 50 output option. Seems like “a simple matter of software”. That’ll work just dandy to a Class 8 or Class 10 SDHC card… but I guess that might eat into Panasonic’s P2 market. Can’t have that 🙂

-Dave