Rainbow-like Moire?

alpi69 said:
I have also promised to shoot some snow, which is comparable to overexposed waves.

I simply cannot see any green pixel garbage.
Click to expand...

Isn't this on higher contrast edges? I don't think your snow shot has any. The darker areas are pretty light to me. But, given your other shots are already testing similar scenarios and you aren't seeing the problem to the same extend in even 50i I am suspect of the camera itself. That difference between 50i and 25p is just so big.
 
huski said:
well, I got the overlap idea from here: http://www.cambridgeincolour.com/tutorials/camera-sensors.htm
Click to expand...

If you look through the Canon literature, you'll see this is not how they do it on the C300 because it doesn't use a Bayer pattern in the traditional sense. Each 4 pixels by themselves generates an RGB value. This is why there was a lot of debate on if it was actually 4K (technically quadHD) or not given how Canon is reading out the information.

On other cameras with a Bayer pattern, they can do the type of overlapped mathematics on determining the colors of the final image.
 
I had a little while to look over previous shots I thought were perfect and pixel peep.

One thing for sure... the extreme detail of this camera (even at -12) can cause aliasing. It is a zoom up pixel peep on the tiny details kind of thing, but it does happen. (Moire, that I still haven't been able to make happen in anything but the most unobtrusive and some would even say practically non-visible manner). I saw this before I even bought the camera on extreme details.. .horizontals mostly, where other cameras couldn't even resolve the info! Turning down sharpness even further will mitigate that, and not really negatively effect the image quality. CP8 (Cinema) defaults to -12. I think I may turn it as low as it goes... what is that -14?

So I believe what we are seeing here is aliasing, pure and simple, combined with some chromatic aberration. The C300 handles this far nicer than the FS100 did. The FS100 frankly, has this problem much worse. That cam's 4:2:0 encoding, combined with a bloomy sensor (F3 does it too; I tested), combined with some aliasing often made for a pink and purple pixelated mess on blown out highlights against darker pixels... period. Even if there was no aliasing and the image resolved ok, I still saw tons of pink and purple on specular highlights, even when I under exposed. I wanted some nice roll off, and less of that mess, which is one of the big reasons I sold the camera.

To see anything like that issue on a camera almost 4x the cost really bothered me at first when I started to look for it. It shouldn't be there. But the more I checked the footage, the more I think all we're seeing is aliasing of a CA halo. I mean, it's almost always just one pixel wide or deep. It only exists on very specific sharp edges, and then when very blown out, and in focus. And even then, it doesn't happen all the time.

95% of the shots I went back and checked with blown to white highlights in the background with dark foregrounds looked fine. They must have not been aliasing, hence looking great even if far off the charts. In a couple of shots that had a deep depth of field, if I zoomed up, I DID see the green/purple aliasing on the edges, once I looked for it. But frankly, once I zoomed back out, I didn't notice it unless I stared for it. It basically looks to the untrained eye like what lenses do (CA) and nothing like the original poster, which I have never been able to reproduce, even in the worst case. I also haven't seen anyone's sample footage here look like that on RedUser. As we all know, shooting water seems fine.

As for what Paul Joy was seeing, I think that was a CA problem combined with some odd 50i artifacting and the color aliasing all stacking up to make a mess. An unusual case. The other examples from other people, I really don't know how they looked so bad. I'm curious to see if those were shot on C-log or not. If you shoot a frame full of this kind of subject, it may be noticeable in motion. This I found in practice is a rare case. But something to look out for, I guess.

The camera does indeed show color aliasing where you have a straight or angular dark surface back lit by a blown out light source. I guarantee if you shoot an easily aliased surface (like fine cobblestones) and blow them out and look carefully.... you'll see some purple and green fringing. It looks like CA (which it probably is in many cases), but the camera has a little trouble encoding it somehow and it shows as green and purple blocks of one pixel in size. It's nowhere close to as bad as the FS100, and people tell me they can't even see it on that camera. But it IS there, and as we have seen, nearly everyone here can reproduce it if they really try and they really look.

I'm going to shoot at 50i to test the theory that is making things far worse (this weekend, when I have time). But the bottom line here is, I think, yes this camera can have color aliasing in some specific instances, just as it can show luma aliasing. No, 99% of people will never notice it; it's almost never more than 1 pixel wide or tall and when watched at 100%, even on a large movie screen, it blends into the background. (This is kind of proven by the fact that the issue did exist in some footage I had shot in the past weeks, but I never actually saw it, even while staring at it on the theater screen before. I had to seek it out. And it didn't dwell in the frame or anything so I really just never saw it. And for the 1% that do see it it's easy enough to remove in post.

A little troubling, but far better controlled than I have seen on other video cameras. I watched Contagion, shot on a Red, and saw color fringing there too. CMOS isn't film.

Bottom line... I think it's wise to bring to Canon's attention. But aside from turning down sharpness, which we all should be doing anyway as this camera has plenty to spare, I don't see this as any kind of defect or problem that will ruin shots. I think we figured out what is going on here.

There's no problem, as Nuno says. Practically speaking, I would agree, for most shooting scenarios. Just turn that sharpness down. :)
 
Rick Burnett said:
If you look through the Canon literature, you'll see this is not how they do it on the C300 because it doesn't use a Bayer pattern in the traditional sense. Each 4 pixels by themselves generates an RGB value. This is why there was a lot of debate on if it was actually 4K (technically quadHD) or not given how Canon is reading out the information.

On other cameras with a Bayer pattern, they can do the type of overlapped mathematics on determining the colors of the final image.
Click to expand...

yes, the c300 does it better, like this: http://www.filmmakermagazine.com/news/wp-content/uploads/2012/01/componentcolorfilters-375x275.jpg
But if you trying to get a 1080p image from a 1080p Bayer sensor, you will interpolate like on my animation, not? If not, I'll correct it:)
thanks
 
huski said:
yes, the c300 does it better, like this: http://www.filmmakermagazine.com/news/wp-content/uploads/2012/01/componentcolorfilters-375x275.jpg
But if you trying to get a 1080p image from a 1080p Bayer sensor, you will interpolate like on my animation, not? If not, I'll correct it:)
thanks
Click to expand...

It's not a 1080p bayer sensor, it's a quadHD bayer sensor. The filter pattern is the same as other cameras, but the demosaic is different. With other cameras, say like the Red, they use methods like you've referenced, where interpolation is used to build an image at the sensor resolution.

This is not the case with the C300. The sensor has 4 times as many pixels, but, they've decided to not do a normal demosaic. Instead, they've decided to take that 4x 1080p resolution and use 4 pixels by themselves to generate a pixel value in the final image.

It's just a different way of doing it. I also think it's debatable if it's better. You lose the cross interpolation ability like your example. Though, many have argued, at 1080p, does that matter.
 
Rick Burnett said:
It's not a 1080p bayer sensor, it's a quadHD bayer sensor. The filter pattern is the same as other cameras, but the demosaic is different. With other cameras, say like the Red, they use methods like you've referenced, where interpolation is used to build an image at the sensor resolution.

This is not the case with the C300. The sensor has 4 times as many pixels, but, they've decided to not do a normal demosaic. Instead, they've decided to take that 4x 1080p resolution and use 4 pixels by themselves to generate a pixel value in the final image.

It's just a different way of doing it. I also think it's debatable if it's better. You lose the cross interpolation ability like your example. Though, many have argued, at 1080p, does that matter.
Click to expand...

:)
exact 4x = less processor intensive = smaller, lighter, faster, less heat. Yay canon!
 
Last edited:
yes, I know, but the animation on the site is about "all the Bayer cameras", and after that I write how the C300 does its thing.
We're telling the same thing I think:)
 
huski said:
yes, I know, but the animation on the site is about "all the Bayer cameras", and after that I write how the C300 does its thing.
We're telling the same thing I think:)
Click to expand...

True, I do see that now, I was just trying to help you for clarity because the way you've written it, honestly, it's very easy to missinterpret. Obviously, because i did, but maybe I was being slow. :)

What is interesting is I've been thinking about this a lot lately after reading your blog. Given each pixel outputs an R, G and B value, if you were to take the G and just assume the same way for both G bayer pixels (even if it is not), you could build a 4k bayer image in post (although at this point it is post curves so I haven't thought about that). You can do this because there is no cross interpolation (unless of course Canon is doing some magic in image processing). So basically, you could build a 4k image where the green resolution would be lower, since there is no way to determine what each of the two green bayer pixels saw, but it would allow you to experiment with debayering in post.

Gonna have to think about this for awhile. :)
 
Hate to resseruct this but the issue is indeed still happening. I called it a non-issue before but sometimes it can crop up and ruin shots. Most of the time it appears you need to look for it. Sure enough, many C300 videos I have seen online show the issue at one time or another in their running time if you look for hard contrast junctions or bright specular highlights. Take a look at this, the railing in the center particular is a mess. It looks worse when moving. And there's nothing wrong with the exposure of this shot.

Canon in emails has to me acknowledged they see something and has sent it to engineering, but hasn't followed up. I'll be talking with them at NAB for sure. This isn't about using "fill light". It's a problem with the camera. One that can often be avoided and worked around, but in some cases, you can't.

PS to an earlier post - the Cinema preset in the camera with Canon Log (not Cinema Lock) has sharpness turned all the way down. And this kind of thing still pops up.

c300_error.png


close_c300_error.jpg
 
And some more examples - this is from about 34:00 on this Alexa vs C300 test (quite interesting by the way... Alexa wins but that's expected). Look around 34:00, and download the original file to see it.

earring crop.png


 
While this isn't something that's popped up in any of the footage I've shot in the 20+ hours I've rolled, I do find the issue pretty easy to replicate. Granted, I usually have to zoom in at least 200% to see it. Thanks for keeping us updated, and I hope Canon is taking this issue seriously. I'd love to see a firmware upgrade that would mitigate the problem.
 
Last edited:
Some of those shots are chromatic aberration. Because the codec is 4:2:2 those will jump out more than if it was 4:4:4. The fs100 does the same but even more as it's 4:2:0. I would BET that the shot of the earrings would look exactly the same on the F3 (recording internally at 4:2:0) with the same lens and lighting.

But some of those shots just look like regular moire. I wonder what's up with those?
 
As an observer. I think that the shots where the fringing is happening tend to be very very high contrast where as a lot of the shots being shown here are just high. The branches/socks.guitar against sky show the lens CA whereas the sun spots glinting on the ocean are massively over exposed because it's the power of the sun in a small tiny area. So strong backlit scenes will be more prone. I'm pretty sure Rick is right in that it's the way the highlights are being handled and clipped that's causing the issue. But then what is surprising is that if it was a debayer problem at 4K why is it being seen when that 4K is down sampled to 1080p?

cheers
Paul
 
You must log in or register to reply here.
Back
Top