Noise and the Matrix

Wow, vimeo finally finished it. Took it close to 12 hours to upload and convert... Damn members.

A subtle look at the color tone difference in the four matrix modes.

Ha! Funny video! Thanks for posting the test. I agree with everyone else in that I like Norm 2 the most. What were the other settings?
 
Hey Dino, I'm curious if the banding is reduced when the Detail settings are left at zero. I find that with 8bit a little noise and dithering can hide the banding pretty good. It's all about happy mediums.

Have done more test shooting with Norm 2, I've decided that chroma should go back to 0 and master ped at 0 vs. -2 on both. I've reviewed footage on Flanders 17", MacBook Pro via AVCCAM Viewer and 100" projection via BD player. Norm 2 matrix has really made a big difference on all displays--not noise free, but very tolerable. Now the biggest issue I have with the image is when highlights are blown out, there is a ring around the object with a chroma clipped color, even if the highlight itself is white. This chroma clipping look is the number one giveaway(besides noise) of a low cost video camera, IMO. I've seen even the HPX500 go yellow or orange on overexposed objects that I don't see with Panasonic AJ series.

As I've told clients for awhile, if you are shooting in controlled lighting environments, use the 170/200A/500(and now AF100 as well), if there is no light control, that's when you really want the P2 Varicam or other Panasonic model with Film-Rec gamma and 600% Dynamic Range. The processing sophistication of the AJ cameras, plus AVC-Intra is what the extra money buys.

Edit: The other issue that's been bothering me is banding. I thought it was maybe the codec, but it's there in live E to E via HD SDI and on Ki Pro ProRes 422 recording on multiple display devices. I did a side by side spotlight test with the 170 and the banding was present as well--not unusual for 8-bit digital displays--but it's a bit more pronounced on the AF100. Regardless, at least AVCCAM is ruled out.
 
Hey Dino, I'm curious if the banding is reduced when the Detail settings are left at zero. I find that with 8bit a little noise and dithering can hide the banding pretty good. It's all about happy mediums.

This would be a good test. I would need a very clear blue sky to give this test a whirl and shoot it at different detail settings.
 
Ha! Funny video! Thanks for posting the test. I agree with everyone else in that I like Norm 2 the most. What were the other settings?

The test was done with
Detail -6
Vdetail -7
Coring +2
Chroma 0
Gamma CineD
Matrix All of them

I'm not going to be keeping Norm2 for Matrix and bump up the saturation by 2. Still not 100% sold on the sharpness. I feel like the image can get sharper without looking digital. I want me some detail and latitude :)
 
You should be able to push the detail up to at least -4 without seeing noticeable haloing. And depending on the shot, if "sharpness" is most important, crank up that detail. If noise starts to become an issue, use the coring to counteract it.
 
Yeah, so is coring like counter sharpness?

I.e. turning sharpness down in camera removes the digital sharpness added to the image, and so coring takes detail away? I would love to read up a little bit on this, going to google it.
 
Hey Dino, I'm curious if the banding is reduced when the Detail settings are left at zero. I find that with 8bit a little noise and dithering can hide the banding pretty good. It's all about happy mediums.

Good question. I haven't tried 0 detail, but I would have thought more detail might delineate the tonal differences, exaggerating the banding. I'll give it a try. I used a 1K Arri fresnel in full spot on a gray flat field to test the banding. I overexposed and looked at the image in 3200K and 5600K, the latter making it easier to see the banding.
 
Yeah, so is coring like counter sharpness?

I.e. turning sharpness down in camera removes the digital sharpness added to the image, and so coring takes detail away? I would love to read up a little bit on this, going to google it.

Coring primarily affects the 0 to 20 IRE range, where noise is most visible. Raising coring level reduces enhancement in these lower frequency/IRE areas, without affecting high frequency enhancement too much. Back in the prehistoric days of Ikegami HL-79's, they called it "Threshold". Ikegami cameras were always quieter in the blacks and mid-blacks than Sony cameras, but they were softer in that range as well, due to a more aggressive coring setting. They saw no reason to enhance blacks. It is very easy to see this using a chip chart. Basically, we would set coring so that the third chip from black would show image enhancement, trying to avoid any detail overshoot below that chip. The Sony's seemed to show detail in the lowest levels no matter what the settings, resulting in noise. Ikegaimi camera looked more filmic even before the term was coined! Sorry for the '80's flashback.
 
Yeah, so is coring like counter sharpness?

I.e. turning sharpness down in camera removes the digital sharpness added to the image, and so coring takes detail away? I would love to read up a little bit on this, going to google it.

I think coring is actually noise reduction in Panasonic-ese. Someone else can chime in.

But, great test. Made me smile too.
 
The coring is a setting that sets a threshold beyond which the edge enhancement effect won't try sharpening things. The camera sees little tiny fine details and the edge enhancement circuitry goes in and tries to sharpen those up, exaggerating the contrast in them, to make them more visible. Which is fine and dandy, except that the camera can't tell the difference between REAL fine detail, and noise (which looks like little speckly fine detail). So it'll go merrily along sharpening up and enhancing the contrast on the noise, making it massively more visible.

Coring lets you raise or lower the threshold of which the edge enhancement will work. The higher you set it, the more fine detail it will exempt from edge enhancement. The lower you set it, the less detail will be exempt -- and, therefore, the more that will get edge-enhanced. So the goal would be to keep as much of the "good" sharpening that you can, while avoiding sharpening the noise. Coring helps that by letting you set the level as low as you need, or raising it up to get rid of the appearance of noise. However, coring can't tell the difference between fine detail and noise either, so if you raise the coring too high you won't get any sharpening on your legitimate fine detail -- and that can lead to things being too smooth or even "plasticky". So in an ideal world you'd want to set the detail as high as you can stand, and the coring as low as you can stand, but noise gets in the way of that, so you have to do a tricky little dance of lowering the detail and raising the coring to filter out the noise while keeping as much sharpness as you want.
 
The coring is a setting that sets a threshold beyond which the edge enhancement effect won't try sharpening things. The camera sees little tiny fine details and the edge enhancement circuitry goes in and tries to sharpen those up, exaggerating the contrast in them, to make them more visible. Which is fine and dandy, except that the camera can't tell the difference between REAL fine detail, and noise (which looks like little speckly fine detail). So it'll go merrily along sharpening up and enhancing the contrast on the noise, making it massively more visible.

Coring lets you raise or lower the threshold of which the edge enhancement will work. The higher you set it, the more fine detail it will exempt from edge enhancement. The lower you set it, the less detail will be exempt -- and, therefore, the more that will get edge-enhanced. So the goal would be to keep as much of the "good" sharpening that you can, while avoiding sharpening the noise. Coring helps that by letting you set the level as low as you need, or raising it up to get rid of the appearance of noise. However, coring can't tell the difference between fine detail and noise either, so if you raise the coring too high you won't get any sharpening on your legitimate fine detail -- and that can lead to things being too smooth or even "plasticky". So in an ideal world you'd want to set the detail as high as you can stand, and the coring as low as you can stand, but noise gets in the way of that, so you have to do a tricky little dance of lowering the detail and raising the coring to filter out the noise while keeping as much sharpness as you want.

awesome, both of you thanks for the explanation.

I know there is never a magic solution to the perfect setting, because then there wouldn't be any feature to modify, but have we come close to a prefered setting? I see Coring +2 seems to be pretty popular and then it's just a quesiton of detail.
 
Hey Dino, I'm curious if the banding is reduced when the Detail settings are left at zero. I find that with 8bit a little noise and dithering can hide the banding pretty good. It's all about happy mediums.
Yes, the cleaner the signal, the more likely that you'll see banding, and yes, a little noise helps prevent banding. Canon found this out the hard way on the XL1; they were using a very low-noise chipset and they found that there was significant 8-bit banding in the image, and their recommendation to the end users was actually to put on +3dB of gain(!)

The banding won't be visible on a live monitor, it happens after compression; and the more-compressed the signal, the more likely it is to happen. At 24mbps it probably won't be too noticeable, but if you put the camera in 6mbps mode you'll probably see massive bands. The compressor sees big swaths of similar-to-almost-identical image, and rounds it all off to be "the same", and that becomes one big band. Then when you move into a range of slightly-different exposure, it'll smooth that all out into "the same" and that becomes another band. A little noise in the signal prevents the compressor from rounding all the sections off to being "the same", and keeps the banding at bay.

The real solution, of course, is to have a 10-bit or 12-bit codec with a huge fat bitrate. But until that happens, this is an artifact that can happen in 8-bit codecs, it's been with us since the first days of digital video, and it'll be with us as long as 8-bit exists as a recording medium. A little bit of noise is the antidote to banding.
 
I know there is never a magic solution to the perfect setting, because then there wouldn't be any feature to modify, but have we come close to a prefered setting? I see Coring +2 seems to be pretty popular and then it's just a quesiton of detail.
Again, it all depends on what you're trying to accomplish. All those settings are there for a reason. And keep in mind, coring is dependent on your detail setting; the lower the detail level, the less the edge enhancement will be doing, and the less the coring will be able to affect. The higher the detail level, the more the edge enhancement does, and therefore the effects of coring are much, much more visible. So you can't just say "+2 is the good setting" because +2 looks very different when detail is -7, than it does when detail is +7. Like I said, it's all a balancing act.
 
Barry, what is the translation of the Coring setting? Is it specific IRE levels as Dino said? And if so, what IRE is each Coring increment? It would be great to know so can better understand where to look for the changes in the image in our tests.
 
It's fairly obvious, and not limited to any specific IRE value; it's just "more" or "less" as far as I know. The easiest way to test it is to set up two identical scene files, one with value "A" and one with value "B", and then just toggle between them. That way you can easily see -7 vs. +7, etc.
 
Crispening and Level Dependency is another term for Coring or Threshold. Here is another stab at a definition of coring I saw on CML:

Coring is used to remove fine detail information that does not contribute significantly to the detail of the picture but which adds noise to the image. Imagine the detail information viewed on a scope. About the baseline you'd see primarily the noise information, with the detail extending beyond that. Now imagine that you sliced (or cored) this signal so that only the information above the noise on the baseline came through. You would be left with most of the detail information intact but with much of the noise information removed. The coring adjustment determines how far from the baseline the detail information removed. You want to use just enough coring to reduce the noise in the picture but not so much that the fine detail in the image is affected.

And another:

The definition for coring/crispening/level dependency/threshold are sometimes interchanged, even by some camera manufacturers. Generally though, what you described is usually referred to as Level Dependency or Threshold - detail correction that is gated on and off by the level of the video signal itself. This affects the entire detail signal. Coring or Crispening is the process of removing low level detail information (think of it as removing the core of an apple and leaving the important parts of the apple intact) irrespective of the level of the video signal itself. This affects only the low amplitude portions of the video signal.
 
I thought about buying one of those Du Monde charts, but I see they're about $700+. Any recommendations of alternatives that are much, much less money? I'd like to do some of these tests on my own to see what I come up with.
 
Back
Top