Aliasing

Really excellent article. As a 5d owner, this is exactly what I really didn't want to believe, so thanks for laying out the incontrovertible evidence and forcing us all to face facts. It does make me want to find a good res chart and test it on the 5d, though I suspect its real resolution is only slightly better than the 7d.
 
Last edited:
Barry, I think this article does a lot to quell the hatred you sometimes get (from fanboys, etc.).

The key for me is to point out that it may not matter to the user. For some, the picture looking good is all that matters, regardless of how it got there. It seems that for some people they need you "permission" to think that way. The usual conversation goes:

"This is what I'm seeing with tests: 500 lines of resolution"
"No! It looks so much nicer than my EX1, way more sharp!"
"Yes, it's tricking you, the EX1 resolves more lines"
"But...I like the way it looks"
"Okay, it's still not really sharper"
"Just tell me it's okay!"


I like that in this article you put more information about what the facts mean for the user.
 
Thanks. My intent with these articles is to point out what happens, so that folks can be prepared to combat or avoid troublesome scenarios, and will know the limitations that these new technologies provide.

Article has been updated with a couple of "real world" examples where aliasing contaminated and ruined shots, even in natural settings.
 
Actually, the 5d aliasing examples that you posted are quite mild. You should see some of the shots that don't get posted to vimeo. Twilight beach scenes are notorious.

On the other hand, the 5d can take twilight beach scenes that will knock your socks off, if you have the time to hit some shots and miss others. To my eye the 5d's rich color in low light is heads above any other prosumer camera (<$10k). So is its depth of field, of course. I don't agree that other cameras are equivalent "with proper lighting". Being able to shoot with less light opens up creative possibilities. What you call "proper lighting" is likely to be in my mind "the best compromise between the lighting I would prefer and the lighting the camera needs". Some cameras give you a wider range of lighting possibilities than others, and that's a definite advantage in terms of artistic range, and not just convenience.

You're absolutely right that all DSLRs have unprofessional sound, but that doesn't exclude them from professional use. It just means you need an external recorder. The internal audio is fine for home and most web video, and quite handy for manual syncing with external audio.

I also agree that it would be foolish to use a DSLR as primary video camera at a live event for a paying client. Unplanned situations force you to use a wide depth of field, which hugely increases the likelhood of moire, and nullifies one of the DSLR's main advantages.

However, you can do a very professional job with the 5d as primary camera for independent TV and cinematic work. It requires more pre-shooting and botches more shots than a pro crew and pro actors would tolerate, but for self-produced, no-budget projects, it's one of the best options. Still, it's probably best used in tandem with a standard videocam.

My wife and I have been shooting with the 5d for almost a year now, and I think we've run into just about every flaw it has. Until I read this article, I wasn't sure if some of these flaws were the result of aliasing or the codec, so thanks for solving that question and laying out the case in favor of aliasing being the primary culprit. But there's one more thing that I'm still unsure about.

To my eye, compared to conventional video, the 5d often looks flatter. By that I mean the 5d's image is somehow less convincing in producing the illusion of 3d. This is a very subjective judgment, and maybe others will disagree. But I'm wondering if that also could be a result of aliasing. Could aliased fake details be stomping on the subtle real details that give the illusion of depth?
 
I think that perhaps what you're seeing is that actually, even when sharply focussed, the 5D2 is quite soft. Resolution measures very low for a "1080p" camera, and standardly aggressive edge sharpening is applied. That means there's very little in the way of real micro-detail, and what is left gets squashed by the codec. I reckon that could quite easily account for the lack of "3d effect".
 
Thank you Barry for that great article. I'm a bit late on this discussion so I probably will repeat things that are already said. I own a 5D and I'm conscious of its limitations. But I always stated that I'm not too keen on sharpness. I sympathized for a long while with the HPX 500 and though we are talking about totally different cameras, I would say if you can live with a resolution of 720 lines (or less) the 500 and those video DSLRs are great tools. I really love their low light, DOF and color capacity - the only really ugly and distracting thing is aliasing and when one uses the now commonly known tricks one will get a pleasantly looking 720p image. Within one or two years we will have full HD resolution - but for me that is not the quantum leap that represented this first generation of video DSLRs, It's evolution. It is welcomed, of course, but is not a game changer anymore. Consider your video DSLR as a 720p camera and you will be happy. I am ;)
 
Barry_Green said:
Problem is, if you eliminate all the aliasing, you're going to be left with only the true resolved detail that these cameras can provide, and that's not much more more than standard-def. It is the presence of aliasing that makes these cameras look sharp at all.
Click to expand...

This is a great article Barry, and important info for folks to have if they're planning on working with these cameras.

You categorize the rolling shutter effect as being a kind of aliasing. That's the first time I've heard it described this way. The classic wagon-wheel example makes sense as aliasing — the sample rate is two low to accurately depict what's going on, rendering the impression of false information. But I'm not quite seeing how rolling shutter adheres to this definition.

So you threw me there at the beginning, but then you grabbed me right back again. I think I often confuse people when I write about the low effective resolution of the 5D and 7D, now I have a reference to point them to. Thanks!

-Stu
 
The idea with the "rolling shutter as aliasing" is that the slow sampling rate of the rolling shutter is not letting the system always accurately represent the information it's attempting to. Look at the "bendy propeller" and you might see how it's kind of the same situation as the wagon wheel, but driven by the slow progression of the rolling shutter down the frame. It misrepresents the propellor as actually being horizontal instead of vertical. It's a "motion" version of the res chart extraction I pulled out, where mainly horizontal lines are actually represented as primarily vertical lines. Might be a little bit of a stretch, but the main goal I was going for is that aliasing results in image (or sound or motion) representations that are not accurate. And the rolling shutter bendy propeller certainly did that.
 
I see rolling shutter as a seperate issue from aliasing. There is currently a passionate (perhaps even overly-passionate) discussion about LCD shutters over at Cinema5D with a member claiming that one is on the market in Paris for the 5D. There is also a vimeo of this kind of solution as proof-of-concept here:

http://www.vimeo.com/5976527

Aliasing remains for me an artifact of accentuation, whereas rolling shutter is the outcome of the slow reading-out of the sensor. Might be good to have a seperate article about rolling shutter if there isn't one already. (Come to think of it, I think there already is somewhere here...)
 
yoclay said:
I see rolling shutter as a seperate issue from aliasing.
Click to expand...

I think you're right. It's the only kink in an otherwise perfect article. I just linked to it on Twitter and have seen about fifty retweets.

Barry, I would humbly suggest removing the rolling shutter reference. The wagon wheel example is a better on-ramp into the discussion.

-Stu
 
ProLost said:
I think you're right. It's the only kink in an otherwise perfect article. I just linked to it on Twitter and have seen about fifty retweets.

Barry, I would humbly suggest removing the rolling shutter reference. The wagon wheel example is a better on-ramp into the discussion.

-Stu
Click to expand...

Though it is clear that the limited read out speed of the sensor is the (main) cause for both problems.
 
from the article:

For some reason, though, people want to give aliasing a “pass” when it comes to the new still cameras.
Click to expand...

This brings us back to the wagon wheel - for "some reason", people have given aliasing a "pass" when it comes to 24fps for quite a while now. Actually, lately, not just given a pass but actively demanded 24fps despite the temporal aliasing. I would suspect it's the same reason the HDSLRs get a pass on their spatial aliasing - the overall aesthetic they produce is worth the trouble of shooting around (or simply accepting) the aliasing. The audience has been watching wagon wheels go backward for a lot of years without complaint - I suspect the the same will be true for the spatial aliasing of these cameras as long as your story is engaging.
 
I agree. 24fps is clearly not able to accurately resolve motion as well as 60fps, and the reason is the same -- lower sampling accuracy. 24 is less than 60, and 60 samples result in much more accurate motion rendition.

BUT -- we (the moviemaking/moviewatching public) LIKE the look of 24fps (in most things; it can get headache-inducing if movement isn't controlled). I don't think anyone actually likes the wagon wheel effect, but we put up with it for the rest of the benefits.

The higher the sampling accuracy, the more accurate the representation of the thing being sampled. But in film, that's pretty much what we don't want -- we like the "surreal" or "larger than life" or "dream state" look of 24fps motion, instead of the "hyper-reality" look of 60fps.

The DSLRs certainly use aliasing to "punch above their weight class", giving images that, in the right circumstances, look like they came from far more expensive cameras. Without the aliasing, they'd never create objectionable image artifacts, but they'd also be substantially lower "sharpness".*

It is a tradeoff that you need to know you're making when you get into the game. The aliased DSLR look is sometimes incredible, and sometimes it can cause distracting or even shot-ruining artifacts. But at the price point, I think it's probably the right compromise, as long as people know what the compromise is, know how it manifests itself, and are on the lookout for it.

*the "resolution" would stay exactly the same without the aliasing; aliasing isn't resolution, it's an artifact. What would disappear is the "false sharpness". Which is what gives the DSLRs their "punch".
 
Barry, your article is making it's way through the internet; ProLost, slashcam.de, canonrumors.com...it's becoming a reference :)

There are some thoughts running round my head, I'd be curious about your opinion.

There are common modifications for astronomy- and infra-red photography, where the original anti-aliasing glass in front of the DSLR sensor is removed and replaced by other (special) glasses for astronomy or IR, or just normal glass or UV-filter glass. All replacements result in much sharper images and higher light-sensibility of the sensor due to the lack of the anti-aliasing/lowpass-filter glass.

http://www.maxmax.com/hot_rod_visible.htm

http://www.baader-planetarium.de/sektion/s45/s45.htm

http://www.optik-makario.de/
(sorry, last 2 are german)


Now, the caprock anti-moire filter seems to work nice on the 7D,

http://www.dvxuser.com/V6/showthread.php?t=189857&highlight=caprock

so I'm wondering if it would do the job on a modified 7D in video mode as well? I'm very insecure about the whole issue, so I could imagine any result from becoming absolutely useless for video to being "real" sharp with less moire if the caprock filter can deal with the new amount of aliasing. The modification seems to work great for stills photography but I understand video and the meaning of anti-aliasing for video is a whole different thing and I'm not sure what role the DSLR's software plays in this.

EDIT: I'm very sorry if this has been discussed here already, I didn't find anything yet though
 
Last edited:
All replacements result in much sharper images and higher light-sensibility of the sensor due to the lack of the anti-aliasing/lowpass-filter glass.
Click to expand...
But "much sharper" is due to the exact same problems we've been discussing in this article. Removing the anti-aliasing filter means that you're letting all the aliasing through, which means you'll have all the moire and jaggy line problems and false detail that we've been talking about.

So while it may appear to be sharper, it's due to cheating and fake detail. Whether you (or any viewer) prefers that look is, of course, a matter of personal preference. But it's not actually sharper and it's not actually recording any more accurate of a picture, it's actually recording spurious inaccurate detail and presenting it as if it's accurate.

Now, the caprock anti-moire filter seems to work nice on the 7D,
Click to expand...
The Caprock filters can certainly overcome all of the aliasing issues, turning the DSLRs into properly anti-aliased video cameras. But, as I said in the article, you'll also be giving up all the fake detail that makes the images look so "sharp" in the first place! You'll be left with a camera that delivers somewhere around a reasonable 720p image, because that's all that the current crop of DSLRs (5D/GH1/7D) can deliver. Anything beyond that, is aliased false detail. The caprock eliminates the aliasing, which means that the false detail goes, leaving only the actual resolved detail. On these SLRs, that means about 600 lines.

Now, that's not bad, and I'm actually looking at picking up some of those to really experiment with. It would remove any uncertainty about the SLR's image, any worries about moire or rainbow patterns, etc... and still deliver sharpness around HVX200 level. With shallow DOF. For a grand total of about $2500 (camera plus filters). That's not bad. The Caprock filters need to be tested in combination with the SLR and the lenses, because each focal length lens will need a different strength of anti-moire filter.

Of course, a much less expensive way to go, that would work with every focal length lens, would be to just de-focus very slightly. :)
 
Barry_Green said:
Removing the anti-aliasing filter means that you're letting all the aliasing through, which means you'll have all the moire and jaggy line problems and false detail that we've been talking about.
Click to expand...

Which I thought would be good when using the Caprock filter because:


Barry_Green said:
The Caprock filters can certainly overcome all of the aliasing issues, turning the DSLRs into properly anti-aliased video cameras.
Click to expand...

By removing/replacing the A-A glass and using the Caprock filter instead the image should stay quite "sharp" (as there is no more additional blurring from A-A glass only softening from Caprock filter) but be less "faked" with less moire issues:


Barry_Green said:
But, as I said in the article, you'll also be giving up all the fake detail that makes the images look so "sharp" in the first place!
Click to expand...

I might be really confused here though :)
 
Last edited:
Well... I can see where you're coming from, but I doubt you're going to see any benefit to removing the AA filter from the 7D. The 7D's AA filter is tuned for the stills, and its video resolution is already so low that I bet the 7D's AA filter really has little to no effect on the video footage at all. The Caprock would probably filter out any and all too-fine detail, leaving nothing for the 7D's AA filter to catch. So if you like the look of footage from a filterless camera for stills, but want AA for video, then that's a case where removing the 7D's filter might make sense (and compensating with the caprocks).
 
Barry_Green said:
Well... I can see where you're coming from, but I doubt you're going to see any benefit to removing the AA filter from the 7D. The 7D's AA filter is tuned for the stills, and its video resolution is already so low that I bet the 7D's AA filter really has little to no effect on the video footage at all. The Caprock would probably filter out any and all too-fine detail, leaving nothing for the 7D's AA filter to catch. So if you like the look of footage from a filterless camera for stills, but want AA for video, then that's a case where removing the 7D's filter might make sense (and compensating with the caprocks).
Click to expand...


Thank you Barry!
 
You must log in or register to reply here.
Back
Top