Demystifying Noise

[yevlar]

I know the noise that we're seeing in some (not all) of the posted clips is being caused by various post-processing like the H264 compression, 10-to-8-bit conversion, ad nauseum, and is NOT on the actual footage captured by the Red; but I worry that seeing the noise without an in-depth explanation of the underlying causes might give ammo to those silly "Say-ers of Nay." Plus, I'd kinda like to know the reasons myself. ;-)

So, I was hoping that someone smarter than I could give an in-depth, insanely technical explanation of the causes, perhaps even with before-and-after comparison images, to sort-of demystify the noise and show that it is, in fact, not inherent in the camera. The ideal situation would be for everyone to see it projected in 4K from the raw sources, but not all of us have that kind of access. And perhaps someone could offer up some compression tips to help reduce the occurrance of noise in posted clips.

[nawaf65095]

Good post and i hope that the RedOne camera is not producing the Noise in picture with non lighted locations also i hope that the RedOne camera Noise clear than Sony HDcams picture.:shifty: (not in picture size..i mean in picture clearance in dark places).

Best regards

[yevlar]

The footage that's been posted by Red themselves, the clips from CTL, and the recent Great Salt Flats footage, are pretty much noiseless, and some earlier low-light tests (driving through the garage for one) are very low in noise, so I'm confident that the camera is not producing the noise, at least not at ISO 320 (higher ISO's will naturally show some grain regardless.)

[Kevin Halverson]

I am curious, are you really wanting to asses the camera's performance in, as you say, "...non lighted locations" and do you mean radically under lit or just in scenes that you want to make look unlit?

There is actually a very big difference between these two situations. One of the things that has become apparent is that with the RED ONE's fixed analog gain (at least for now) one is simply trading off the S/N ratio against the minimum illumination level in any given scene. The imager and more importantly the receptor size of the camera give it some distinct advantages, but its not violating any rules of physics (it is perhaps bending a few monetary ones however), noise character and performance will always be bound to these relationships.

Now, on the other hand, you are considering how to achieve low noise in a scene where you want it to look unlit, that is all about good lighting and good post work. This is worth repeating, to look unlit takes excellent lighting.

Keep in mind that it is nearly impossible to fully evaluate any of the noise character from frame grabs and compressed clips alone; so trying to come to conclusions based upon the materials that are being posted may be difficult. You would ideally want access to a few seconds of an uncompressed sequence to have a decent idea of what to expect.

[Nook Kim]

If he is referring to some situations in documentary, ENG, ect., there could
be lots of unlit locations that people need to go through.

[David Mullen ASC]

Noise and exposure are related -- you can't expect any digital camera to deliver the same noise level in wildly different exposure levels. At its "normal" ASA level (320-500 ASA) the RED appears essentially noiseless, with increasing noise levels at higher and higher ASA values, just like you'd expect. If your "unlit" location has enough light for a usuable f-stop at, let's say, 1000 ASA, then you'll probably be fine with the minimal noise level, but if your unlit location needs to be shot at 8000 ASA just to shoot wide-open on the lens, then you'll be seeing more noise.

And there is always the option of noise reduction in post, though not a cure-all.

[Kevin Halverson]

...(higher ISO's will naturally show some grain regardless.)

This isn't exactly true. It isn't the "higher ISO's" that increase "grain" (noise), rather it is the lower illumination levels and the resultant lower S/N ratio. Keep in mind that, at least for now, the analog gain of the camera is fixed. Changing the ISO setting does nothing at all to the image, rather it simple changes the meta data. All that determines noise is the scenes illumination level.

[Chris Nuzzaco]

I've actually been doing many tests recently with my Andromeda with regard to noise, and I have found that various white balance settings under various color temperatures can cause noise levels to change in each channel. Now it is important to consider that my camera is 3 chip, and not bayer filtered single chimp CMOS... but I believe the same principal applies. All your white balance ever does is adjust the gain of the red, green, and blue channels. I'm not sure if red lets the white balance mess with the green channel though. I have found that the DVX100's green channel never changes, only the red and blue.

I've also observed that the blue channel is generally the most noisy, followed by red, and then green. I'm quite surprised at how clean the green channel is under various color temperatures. Sometimes red and blue change positions with regard to visible noise when shooting under different color temperatures, and overall scene color affects this also.

I'm currently getting even cleaner footage right now by white balancing in such a way that I force the image too look somewhat green, and then optically color correcting with various magenta filters. In post, I will only noise reduce the red or blue channels depending on which is more noisy, and the results are nothing short of amazing. Some would argue that noise reducing the footage I'm pulling out the system is overkill, but if you plan to blow up to 2k, every bit of clean footage helps substantially.

Hope that helps!

[Kevin Halverson]

Thanks for your contribution Chris. I know that you have been getting some excellent results with your Andromidazided DVX, however, there is some very big differences between these two cameras.

The DVX (Andromidazided or not) has user adjustable analog gain in two areas. The overall gain of all three channels (RGB) is adjusted by the on camera gain switch and the white balance effects two of the three channels (R & B). In the case of the RED ONE, this isn't the case. Rather, on the RED, changing the ISO or white balance only changes the meta data, not the image that the camera captures. This is very different than how your camera behaves. The only thing that will effect the noise level is the minimum scene illumination and post processing, in camera setup choices will not.

[jbeale]

My understanding is that the RED will show equal R,G,B channel clipping on the sensor with a light of approximately 5000K. That means if you are working with normal tungsten lighting you'd expect to see more blue-channel noise than in regular sunlight. Or in a heavy overcast, high color temp. light you'd expect relatively more red-channel noise. That said, assuming you have enough light of whatever color, the Red noise levels look as good or better than any camera I know of.

[Kevin Halverson]

jbeale, you are correct. Lower CT lighting will slightly underexpose the blue channel and as a result, you would experience a bit more noise due the the decreased S/N ratio of those receptors, the inverse is also true, but I would expect to a lesser degree due to the nature of the CFA. Establishing the S/N performance under different CT conditions is a test that I will be running as soon as I can get my hands on a camera.

[Chris Nuzzaco]

Thanks Khmuse!

From my observations, its probably easier keeping noise levels down on a bayer filtered CMOS chip simply because you are only dealing with one sensor and one gain setting, thus eliminating the variables I have to deal with in the 3 chip world.

I do wonder about channel gain and white balance though, in post, this is where my system can be somewhat similar to RED One. Even though the camera captures with one chip, you still need to either brighten or darken various channels, which can increase visible noise if you really need to crank up one of the channels, albeit, in post.

As for more in depth information on noise, I've heard that some wavelengths of light are simply harder to get a clean capture of using various electronic sensors, so noise isn't always an RGB or channel gain issue, whether at the capture stage, or post processing stage.

http://electronicdesign.com/Articles/Index.cfm?AD=1&AD=1&AD=1&ArticleID=1289

It would be interesting to see RED One footage shot using a magenta filter and then color corrected in post. I wonder if this could eliminate the blue channel noise some people have been seeing in the footage. The technique theoretically increases your dynamic range as well, because you can keep more of your top end (helps decrease uneven channel clipping aka "magenta highlights").

The technique stems from the Viper Filmstream system (3 chip). The footage is green due to the way the channels are set up gain wise, and locked off. The magenta makes the image colors normal looking. I can't be too sure this will work with RED, its just blind guessing, as I don't know all the secret tech specs of the sensor.....

Perhaps the camera is called RED One because the sensor has a red looking filter in front of the Mysterium chip, maybe its actually magenta...

... And maybe the Mysterium chip is designed in such a way that the footage coming in without this filter looks green, thus becoming very similar to the Viper Filmstream, only using one chip. They did say this sensor was custom built for them....

*The last few comments about the Mysterium sensor are PURE SPECULATION!!!!*

LOL

:)

[Kevin Halverson]

Hey Chris,

Thanks for your contributions here. As for your idea, I would think that hitting the optimum CT by either on camera filtration or lighting filtration should yield the same results at least as far as S/N optimization is concerned. To me, the task is to determine absolute natural balance and then establish S/N changes with different CTs. Then you are armed with the knowledge that allows you to weight the variables. You can then make informed lighting decisions and leave very little to chance.

Kevin

[Chris Nuzzaco]

Thanks interesting idea there Kevin. So far I have only done tests using 32k and 56K. Perhaps I need to calibrate the white balance under something in between, using magenta to white balance on (magenta ensures a somewhat even adjustment of blue and red). My best 32K results still show that my blue channel is too sensitive, but I have managed to get the red and green channels really close. The unfortunate thing with Andromeda is that its white balance is a bit unpredictable :(

I'll have to try slapping some light color correction gels in front of my lights during the next round of tests.

[Graeme Nattress]

We have software algorithms in our apps that can deal with any purple nasties in the highlights. Given you can't get a filter that boosts blue, only one that diminishes red and green, you're in the situation where any time you make the image darker, you increase noise. So unless you were going to be using an ND anyway, I don't see a magenta filter helping you much in overall noise, and it complicates post workflow.

Graeme

[Chris Nuzzaco]

I can't speak for Red One, but in the 3 chip realm, many Viper shooters swear by using a magenta filter. Many of those DPs have noticed that correcting the green cast in post always shows more noise post correction as opposed to doing it optically and compensating for the darker image by opening up the aperture, or just bringing in brighter lights. The big overall idea is to simply get all three chips to have equal gain levels, since I use a moded DVX 100, I have to really work at tricking it as best I can, viper is just locked off at the factory.

So far all of my tests show lower noise after I balance on magenta (creating green looking footage), and then filter optically with magenta (to correct the green cast), and compensate for the decrease in exposure (either by opening up the lens or using brighter lights). Whenever I manually balance on white and let it be, the noise increase is very evident, even more so when white balancing under very low color temperatures.

[Kevin Halverson]

In a 3 CCD camera whatever yields the most uniformly high signal on all three imagers would seem like a valid idea if optimizing the S/N ratio is the goal.

[Graeme Nattress]

White balance has zero effect on the recorded R3D file. If you fix the purple highlight, should you get one, in post with our DRX feature, you don't have to gain up darker footage to fix the highlight.

Graeme

[Chris Nuzzaco]

Interesting Graeme, so how does RAW white balancing work? As for Viper, white balance doesn't even exist in Filmstream mode, all the more reason for an optical filter somewhere along the way.

I understand your highlight color correction feature, I used something similar in After Effects a long time ago to fix my highlights when shooting linearly, but what about the rest of the images tones?

[Graeme Nattress]

White balance works by:

the camera measuring (by you pointing it at something white) the correlated colour temperature and distance above or below the plankian locus of the light. These two numbers, kelvin and tint are stored in metadata along with the RAW file, and are also used by the in-camera image processing to produce an image for the viewfinder and outputs etc.

In post, the RAW file is decoded and is still RAW. Then it gets demosaiced to RAW RGB, and then image processed. Image processing takes either the metadata white balance info or the user set white balance info, or you can use a picker to pick the white balance.

Now we know the colour temperature of the light, and we also know the colour temperature of the desired output space, which is usually D65, and we can calculate a mathematical transform to take the measured scene white point to the output white point.

This white balance effects the entire image and corrects the colour for it to balance it. If it, by happenchance, produces an offwhite highlight, this is then corrected with the DRX feature. If you set an off-white white balance, then the DRX corrects through to the off-white white also. It's quite clever.

Graeme

[Chris Nuzzaco]

Very interesting, now when the balancing actually happens, doesn't that mean the three different channels need to be brightened or darkened somewhere along the way? This is what I've been getting at, when you have to brighten up say the blue channel (which is already noisy) for balancing footage shot under 3200K, you should theoretically increase the noise visible in it. Right? Or am I missing something here that only applies to RAW style cameras? At any rate, having a bayer filtered CMOS with a very very low signal noise floor seems like an absolute must since you can't (as far as I know...) adjust gain in each channel when capturing.

Thanks for the info so far!

EDIT: I did some more thinking... I suppose you could just lower the red and green channels in post instead and avoid amping up the blue ;)

[Graeme Nattress]

Well, adjusting gain will just trade you analogue noise for digital noise. There's an optimum position to have things at, but if you x2 the gain digitally and compare to x2 the gain analoguely, you've still got the same source noise at the sensor, it's just getting boosted in two different ways, and they do look different. To say the analogue gain is that much better is, well, perhaps true and perhaps not. I guess it all depends on the image processing and the precision of how it's done.

Yes, you end up gaining or diminishing the RGB channels indepentently, and if you're blue deficient, you do indeed gain up the blue. It's actually more complex than that, although most video cameras don't make it complex - they keep it simple.

In the end, there's lots of variables - and it's the end result that counts. If you fancy taking the time to try the magenta out, please do - it'll be fun to see how it works. I'd not then want to use our white balance though, as it's calibrated for a clear light path with no filters. Be interesting to see the end results.

Graeme

[Kevin Halverson]

Hey Chris, I think you are latching on to the point that I was trying to make about the RED ONE, namely, that nothing done in camera effects the analog gain, either individually or as a whole. So the natural CT balance and clipping point are really fixed. To optimize, you need to work around this fact. The real advantage of this camera is that it has a low enough noise floor and a 12 bit quantizer, that you really can get some incredible results with these fixed characteristics.

As for being able to actually set white balance in camera, I don't see this being practical as it would require resetting the reference each time a different part of the CFA was read. At the most, all that will be practical (I think) is to have access to a global gain setting. I am fairly certain that in camera setup values are used to "calibrate" the sensor to ensure very high unit to unit consistency, but it seems like the sensor really is amazing in terms of raw performance.

To me, this stuff is fascinating!

Kevin

[jbeale]

Very interesting, now when the balancing actually happens, doesn't that mean the three different channels need to be brightened or darkened somewhere along the way?

Regarding R,G,B noise, check out Graeme's screencap here, bottom right:
http://www.reduser.net/forum/showthread.php?t=4904&highlight=imatest

If you're familar with Imatest, you'll recognize the bottom center-right plot labelled "pixel noise" plotted separately for R,G,B as well as luminance. Note that (except for the extreme dark regions, bottom 5 patches or so) the R,G,B plots lie right on top of each other! Only at a target density near OD 3 (10 stops down from the highlights!) does the blue channel become more noisy.

I don't know if Graeme did anything in post to desaturate color or reduce noise, but if not, his lightbox has a light source which is almost *exactly* matched to the camera as far as balancing the channel noise. He mentioned somewhere that it was near 5000 K. http://www.reduser.net/forum/showpost.php?p=90313&postcount=6

[LighthouseMEdia]

I know after seeing some of the Chicago test footage this is a potential issue for me as well, I would like to see further testing to confirm some of these performance issues. I also noticed how unbelievablly hard it is to see grain in the images at much less than 1K but around 1-2K sizes that were screened in Chi-Town it was very noticable at 1000 ASA.

[jbeale]

Definitely more testing is needed all around. My current suspicion is that noise is not a comparative limitation for Red vs. film. In other words, if you had Red and any 35mm film camera shooting in the same lighting conditions with the same lenses and exposure settings and screen the result, my guess is the Red image will have less apparent noise/grain, regardless of what film stock you use. This is just a personal guess so I will look forward to seeing a test of this type.

[Chris Nuzzaco]

If someone can figure out what color temperature results in a truly even balance of all three channels while pointing the camera at a pure white subject filling the frame, then would it make sense to shoot threw correction filters based off of this starting point? In other words you would determine the cameras optimal balance in all three channels and say the camera is really rated for 3200K or 5000K, etc... and then use filters when shooting under daylight, tungsten, etc. to correct it back to the cameras base color temperature rating? By filters, I'm talking various densities of orange and blue, and not magenta.

[David Mullen ASC]

If you're working in tungsten light, which tends to be night interiors, generally you don't have the stop to use a blue correction filter -- the 80A (corrects 3200K to 5500K) loses nearly two stops, so even a halfway correction would lose one-stop. And if you couldn't compensate for that, it begs the question of whether a better-exposed 3200K image would be less noisy than an less-exposed 5000K image even though it was "optimized" in some way for the native color sensitivity of the sensor.

In the case of the Viper, the controversial issue was in regards to using the magenta filter in high-key situations where bright whites might clip at different points in each color channel, so the magenta filter was mainly used in bright daytime conditions, not for low-light night interior conditions. And in that case, daylight shooting, you generally have the stop to use a light-robbing magenta filter. ("Zodiak" did not use the magenta filter for interiors and night scenes.)

But in this case, using a blue filter in tungsten-lit situations, well, it's not very practical. If it actually helped, I could see it being used in chromakey set-ups where noise issues were more critical, but not for your typical interior scene lit with tungsten practicals or firelight and whatnot.

[Chris Nuzzaco]

I would assume anyone trying out the filter approach would compensate for exposure anyways and just plan their lighting around that. I suppose it could be a somewhat overkill measure for your typical shooting with a Red One given its already very low noise floor, but it certainly would be very helpful when shooting anything for special effects work (green screen, etc.)

Come to think of it, I don't know why anyone would bother with blue screen when shooting video, that channel has historically been the noisiest channel. Except for situations where you have lots of green foreground content, I wouldn't recommend blue screen for any digital effects cinematography, unless of course your using a camera that has a very very clean blue channel, which is probably unlikely, it seems to be a problem with all the various digital cinema cameras, as far as I know.

[Graeme Nattress]

It's not just digital sensors where blue is noisiest, but film too. It's basic physics. We've tried to make blue as good as we can though.

Graeme

[Mike Prevette]

This thread got me thinking of something totally looney. Now this is not recommended nor would it be very practical on set, but for "low light" interiors what about throwing some CTB on the lights aimed at the shadow areas of the frame. Essentially bringing up the blue channels level to above the noise floor, and then correcting the blue out later on?

Fucking nutz I know, but I'd love to test it.

_mike

[Chris Nuzzaco]

Actually, thats not as "nutz" as you think, todays color correction programs are extremely powerful. The more crazy aspect would be controlling the blue fill light on set.... Yikes!

[Mike Prevette]

Hence the "not practical" part of my post. But if the scene were a warm lit interior (bedroom, etc) there wouldn't be much blue present naturally so you could just make all the blue go away, and that would take care of most of your spill if you were careful.

[Chris Nuzzaco]

I wonder if you could just use softboxes with egg creates. This sounds like a fun experiment :) I'm always looking for new ways to push digital cinema to the next level... hence my "Andromedized" DVX...

[David Mullen ASC]

I don't think blue fill light is going to help anything much, noise-wise, but slightly blue/cool fill is not a bad idea -- warm shadows tend to go reddish because they are warm + underexposed, so filling in with whiter, or cooler, light to take the redness out of the shadows can be a good idea.

I usually key with tungsten but sometimes fill with 3200K Kino tubes, which I find to look slightly less warm than tungstens.

In very dark night scenes, Caleb Deschanel, ASC used to write about using blue light in the shadows because it seemed to show up on the negative earlier than other colors -- in other words, you could have a shadow that is very dark, more than three-stops underexposed, but if it were blue-ish, you'd see some detail while not thinking about the fill. While not exactly what I'm talking about, you can see the idea here in this night shot in "The Natural":

http://www.davidmullenasc.com/natural1.jpg

[Billy Summers]

another sweet, albeit ghetto, move is using a UV haze cutter to help reduce noise in the red or blue channels...

:wacko:

[Chris Nuzzaco]

Thanks for the tip David :)

Ok, as for noise though...

I just ran a much more accurate test on my whole magenta filter approach (using a 3 chip Andromeda DVX100, this doesn't apply to Red One or other single chip bayer filtered sensors) and the results...

Green channel: Remains constant, as to be expected, this channel is locked off gain wise, but it is also very clean.

Red channel: This channel shows a very visible improvement.

Blue channel: Well... its a bit of a lost cause. I saw a very very minor improvement in the blue channel, but its very hard to see.

The test was shot using a 32K photo flood while shooting a white card. I put two magenta filters in front of the lens (100 ccM total I believe). I balanced thru the filter and then shot my test. I was very careful to make sure my levels were equal for each white balance setting (with filter and without filter). I have to expose higher up even without the magenta filter due to the chips becoming less sensitive. So far this seems to help me out, it might be somewhat overkill, but as David noted earlier, this could really help when doing effects work, and I'll add that this can also help when doing a blow up. I'll post a link to a video as soon as I'm finished working on it.

NOTE: I did keep my UV filter on when shooting both white balances...

I have a few more ideas I'm gonna run some tests on as well.

[Kevin Halverson]

In as much as the goal is to reduce noise, and noise is most prevalent in the blue channel, why not try your test again but cool your source with some 1/2 or 3/4 CTB. This should raise the relative level on the blue channel and further improve the S/N ratio.

[Chris Nuzzaco]

In as much as the goal is to reduce noise, and noise is most prevalent in the blue channel, why not try your test again but cool your source with some 1/2 or 3/4 CTB. This should raise the relative level on the blue channel and further improve the S/N ratio.

I've actually been thinking about that... It's gonna be a late night!

[Kevin Halverson]

Late night, but if you gain some valuable insights, its all worth it. By the way, can you quantify the results you are getting numerically? It might just end up that lighting with a daylight balanced source and a bit of minus green might achieve the best results.

[Chris Nuzzaco]

OK...

I have some semi results.

First, its important to realize my goal is put my red and blue channel gain right where green is, not above, and certainly not below it, creating total equality.

I lit my white card using 3200K corrected for 5600K using gels. I then balanced on white threw my two magenta filters (about 100 ccM by my estimate). I then pulled off the filter to see where the channels stood, as I have done for all my tests. The red channel wasn't dead on like my earlier test using tungsten, so I actually need just a bit more, probably 150 ccM (or three minus green filters instead of two).

The BIG thing I did notice... My red and blue channels were much closer together! Perhaps shooting under 6500K with an appropriate amount of minus green is the magical setup for best noise performance. Can't wait for my local photo shop to open up in the morning...

:)

[Chris Nuzzaco]

Late night, but if you gain some valuable insights, its all worth it. By the way, can you quantify the results you are getting numerically? It might just end up that lighting with a daylight balanced source and a bit of minus green might achieve the best results.

I honestly don't really know how I could quantify these results... Anyone else have suggestions?

[Chris Nuzzaco]

OK... I tried my test again, at least partially, I doubled up my blue gel...

Almost total equality! In my tungsten tests, the red and blue channels never overlapped each other, I now have them halfway overlapping each other. I still need just a hint more magenta filter and maybe another 1/4 CTB gel and I should be right there :)

[Kevin Halverson]

So how much CTB are you using? Is this double half or double full? I don't know if you have a CT meter, but a crude substitute would be to use a DSLR and shoot your white card with a know setting on the camera and then analyze the frame to see how close you are. With a few passes, you should be able to get a pretty close estimate of where you are at.

Keep up the good work, sounds like you are achieving some usable data from these tests.

[David Mullen ASC]

But what's the practical value? -- are you going to put daylight photofloods in your practicals (no pun intended) to get them closer to 5000K?

Generally you can't add blue gel to lights (or switch to HMI's) because you're matching to tungsten practicals in the scene (or firelight, candlelight, etc.) And you can't filter the camera with blue because you need the exposure. And if you light to a higher level to compensate, you may be killing the natural ambient light in the room.

[Chris Nuzzaco]

I'm using two sheets of 1/2 CTB doubled up.

David, I'm not sure yet about the practical value, and yes, I suppose you could put daylight photofloods into practicals if you really wanted to, and I'm the kinda guy who wouldn't mind trying ;) I could argue against people shooting with ISO 100 film too you know... thats pretty low sensitivity, but, some people shoot it anyways because of the low grain. Practical? well, depends on what your after I suppose.

At any rate, even If I can't shoot with a high color temperature light, my tests still show I do gain a significant noise advantage in the red channel shooting threw even just a little bit of magenta (post white balancing threw it of course). You also have to keep in mind that I'm pulling my data out of a DVX100, the sensors of which do not have the lowest noise floor around, and the photosites are not as nice as larger "pro" cameras, so every little trick I can use is helpful.

[Chris Nuzzaco]

Ok, despite the fact that I don't have all the tools I need to hit perfect equality, I decided to go ahead and capture something anyways and look at the channels blown up several times....

There is a very substantial decrease in the blue channel noise!

As for the Red and Green channels, they seem too be just as low in noise, actually they look pretty equal. I would say the blue channel is now pretty close to the red and green noise wise, and this isn't perfect equality gain wise, so I can't wait to get that extra filter in the morning.

Again, I achieved this result by shooting under light that is probably around 6500K - 7000K and white balancing threw a fairly dense magenta filter, and then shooting threw that filter.

[Kevin Halverson]

But what's the practical value? -- are you going to put daylight photofloods in your practicals (no pun intended) to get them closer to 5000K?

I haven't tried using daylight photofloods in practicals, but I have made sleeves of CTB to use with lower power PH series globes and daylight balanced CFLs when I wanted to match the CT of HMIs.

[Chris Nuzzaco]

Ok, I did a very crude test with my DSLR to figure out what color temperature I was working at when I managed to get my channels evenly balanced, its around 12,200 Kelvin. So I guess that means its a super daylight camera LOL. I noticed that even outside on a sunny day, I estimate you would need like somewhere between 1/2 to full CTB , in front of the magenta filters (I estimate you'd need about 5), to make the outside daylight match the camera settings.

OK, so thats quite a light sucker, off the top of my head, all those filters lose you a ton of stops.

At lest I now know...

But its not all in vain. I can still balance and shoot threw a more modest amount of magenta and achieve better results, just not the "uber" results you'd get if you went all the way.

For a camera like this, and I imagine other 3 chip cameras, I would recommend shooting under daylight balanced light and putting two or three magenta filters in front of the camera. White balance threw them, and shoot with them on. You just need to be careful your not making either the red or blue channels clip before green, equality is the key, a little bit higher up is also fine, as its usually much better than if you hadn't used the magenta filters.

[Kevin Halverson]

Yeah, that is a rather high CT to say the least. This is about as far from practical as you can get and still see where you came from!

In the end, it just shows that using the highest practical CT source is always going to yield better noise results. Its pretty intuitive, but good to have some testing to back it up. Thanks for all the effort Chris.

Kevin

[Chris Nuzzaco]

I shot a crude, but effective test today to see if the magenta filter trick would help my dynamic range... it does!

I found that I gain more usable HIGHLIGHT dynamic range after balancing threw, and then shooting with a magenta filter on. So far I'm pulling a full 9.5 ish stops that have NO strange highlight color cast when balancing and shooting under 3200K, which is pretty good considering I'm shooting 10 bit RGB out of a DVX100...

I'm very excited to see how much range I can pull off using this trick when the 12 bit firmware upgrade comes :)

[Steve Sherrick]

Look forward to some video!

Thanks,
Steve

[proaudio4]

Chris, any more on this testing?

So using your ideal magenta arrangement for your DVX100, how many stops would you say you loss?

I understand the use of this technique with controlled lighting when shooting mid light conditions to help noise in shadows and blacks. Under low light
conditions, is the loss of luminosity worth cleaning the red and blue channels?

Thanks for all your time. This has been one interesting thread!

[BASSAM MSSALATIE]

Hi guy's :
Execuse me cause i am coming from non english photgraphy enviorement can you explain me a little bit what was disscused here

Noise and exposure are related -- you can't expect any digital camera to deliver the same noise level in wildly different exposure levels. At its "normal" ASA level (320-500 ASA) the RED appears essentially noiseless, with increasing noise levels at higher and higher ASA values, just like you'd expect. If your "unlit" location has enough light for a usuable f-stop at, let's say, 1000 ASA, then you'll probably be fine with the minimal noise level, but if your unlit location needs to be shot at 8000 ASA just to shoot wide-open on the lens, then you'll be seeing more noise.

And there is always the option of noise reduction in post, though not a cure-all.

if you have some location sample or article could help me

Hi guy's :
Execuse me cause i am coming from non english photgraphy enviorement can you explain a little bit what was disscused here about (ASA) puzzle rates

[Chris Nuzzaco]

Chris, any more on this testing?

So using your ideal magenta arrangement for your DVX100, how many stops would you say you loss?

I understand the use of this technique with controlled lighting when shooting mid light conditions to help noise in shadows and blacks. Under low light
conditions, is the loss of luminosity worth cleaning the red and blue channels?

Thanks for all your time. This has been one interesting thread!

Sorry for the delayed response, I've continued my own personal research into this technique and come to realize a few more interesting things, which do apply to bayer filtered CMOS cameras like RED One, I'll post more in the next few days. It should be very enlightening :)

With regard to shooting Viper using no filtering...

http://www.grassvalley.com/wp/cameras/viper/The_Viper_Exposed.pdf

Apparently Grass Valley now endorses the Magenta + varying degrees of blue filtering. The paper explains why you achieve a lower noise image with greater dynamic range when filtering optically versus electronically (ie. Post color correction).

Thanks,

[Graeme Nattress]

Interesting looking at the viper curves in that paper - they should a real non-linearity that I just don't see when I plot a Red log / log curve. And we have DRX which "fixes" the highlights, which they don't have. I don't think you can directly apply their scheme to RED. But as always, do your own tests.

Graeme

[Chris Nuzzaco]

Thanks for the info Graeme!

It mainly has to do with green video actually. Even though 3 CCD is a different approach from bayer filtered CMOS... if all the channels in a CMOS system are equal in gain, (which in theory, they should be) and not using any additional filtering, the sensor should produce a green image, just like an equal gain 3 chip camera.

Right?

This seems to make some sense, even though the channels are now located on one chip, the gain in all three is now the same, so as far as over all color balance is concerned, you should wind up having a similar result as a 3 chip camera that has all three of its channels equally balanced, even though they are located on different sensors.

Thanks,

[Graeme Nattress]

I don't think this is a 3 chip v 1 chip issue. The non-liearity is probably just down to engineering rather than something inherent in the CMOS v CCD.

As with all "problems" there are often multiple "solutions".

Graeme

[Chris Nuzzaco]

There certainly are multiple solutions! I found that out very early on, though some are better than others, and I think RED nailed it personally ;)

I'm just one those DP's that has an interest in truly understanding how his tools work, thus letting me maximize their performance.

[Kevin Halverson]

Thanks for the info Graeme!

It mainly has to do with green video actually. Even though 3 CCD is a different approach from bayer filtered CMOS... if all the channels in a CMOS system are equal in gain, (which in theory, they should be) and not using any additional filtering, the sensor should produce a green image, just like an equal gain 3 chip camera.

Right?

This seems to make some sense, even though the channels are now located on one chip, the gain in all three is now the same, so as far as over all color balance is concerned, you should wind up having a similar result as a 3 chip camera that has all three of its channels equally balanced, even though they are located on different sensors.

Thanks,

In a single sensor approach, it is highly impractical to have a topology that allows for different gain (analog gain) on different parts of the sensor based upon the CFA. A "3 chip" camera can more easily take this approach as there is a different quantizer (A->D converter) for each "chip".

Finding the natural CT for the camera is something that is of interest as when this is used, it should provide the greatest dynamic range across all three color channels. Accomplishing this with on camera filtration of course presumes that you have adequate illumination levels in the first place (basically replacing ND with spectrally specific filtration). One could also accomplish the same end result with filtration on your lighting instruments but again would need to be used in places of scrims.

[Kevin Halverson]

My testing found that its just a bit higher in CT than the 5000K point mentioned. I found it to be closer to 5300-5400K. Regardless, the noise performance of this camera is simply stunning.

[Graeme Nattress]

I've not tested it "properly" but I'd agree that it's approx 5000k.

Graeme

[Chris Nuzzaco]

I'm curious, how are you determining the "native" color balance of the camera? I suspect that lower "native" color balance *could* be more desirable.

Are you basing it off the color temperature required to be balanced for in order to make all three channels clip at the same point?

Thanks,

[Kevin Halverson]

Are you basing it off the color temperature required to be balanced for in order to make all three channels clip at the same point?

Thanks,

Basically, correct.

[jbeale]

You could change the effective balance by on-camera filtration, that could get you a bit more dynamic range in some situations (depending on your lighting) but also reduced sensitivity.

[Chris Nuzzaco]

The lower "native" color temperature would certainly explain why the Red One has more range without any filtering then, you don't need to offset the red and blue channels quite so much. However, I wonder if a very sleight amount would help, like say 30ccM, which is quite modest really.

[Anders Holck]

It would be interesting to see the differences of a wedge chart lit by 3200k and 5400k sources, and then white balance them both to the same neutrality.

Have you ever tried that graeme?

Or maybe:
Shoot a real subject lit only with kinoflo's with 5400k tubes, then change to the tubes to 3200k and reshoot. Then see how the white balanced files differs.
Just to get an idea of the difference and if it really matters.
(Even though it might tell more about the CRI difference of the tubes)

[Joseph Mastantuono]

My testing found that its just a bit higher in CT than the 5000K point mentioned. I found it to be closer to 5300-5400K. Regardless, the noise performance of this camera is simply stunning.

Agreed, however doesn't it at some point become a physics problem rather than a digtal-sensor/photochemical one? Since light is a logarithmic function rather than a linear one, doesn't there quickly reach a point of no return where the signal:noise ratio get abysmally low, now matter the process?

Not to get too esoteric here but don't cats have intricate mirrors in there eyes to help with this situation, but even at that point very little color actually comes through due to the optics involved? Or is that a biological impediment in terms of rods/cones, and how eyes work.

I am in no way an expert on the subject and got my knowledge of light cribbing the back of a few boxes of crackerjack, so please correct me if I'm wrong.

Isn't most of this discussion somewhat academic? I mean if you have the luxury of compensating for the light lost by slapping a magenta, or swapping bulbs, etc... I'm sure you have the equipment and manpower to throw another light up and get that extra stop of light to make everything super sharp in the first place...

[Graeme Nattress]

Light is not logarithmic, it's linear, but the perception of light is non-linear.

Graeme

[Chris Nuzzaco]

He might be referring to light fall off from a fixture not being linear....?

[chuck colburn]

I believe light "fall off" from any source is the same as sound amplitude. It's the reciprocal of the square.
Twice as far half as bright, half again closer twice as bright.
I think. lol

[Bruce Allen]

Agreed 100%. I've had similar findings. If I'm shooting something with, say, an HV20, it's almost always the blue channel that needs the most "help".

Certainly it doesn't work for every situation, but if you're trying to decide between shooting with tungsten, or shooting with HMIs / Kinos / LEDs, this should influence your choice in favor of the latter. For example, when I shoot effects plates indoors, I'll always go the 5600K route.

Bruce Allen
www.boacinema.com

[Kevin Halverson]

I believe light "fall off" from any source is the same as sound amplitude. It's the reciprocal of the square.
Twice as far half as bright, half again closer twice as bright.
I think. lol

Your calculation is off a bit, at twice the distance, the energy density is 1/4 not 1/2.

In free space, energy density diminishes predictably, however, whenever light or sound (or any form of energy) encounters a boundary, the behavior is modified based upon the reflective properties of the boundary interface.