Recently, Jim Jannard made a post that some people had poor results wtih RED ONE in tungsten light. He recommended using a blue filter and increasing the exposure accordingly.

Last week I shot some stuff in the Washington DC subway system, and even though I custom color balanced, the result is kind of a sickly yellow green. I'm thinking that there's just not enough blue (or magenta) light to produce a pleasing image.

I wonder what the optimum color balance is for the 5D? Maybe, like with RED ONE, it's best to use a blue/magenta filter, color balance with that, and bump the exposure by a couple stops.

Anybody else have sub-par results in certain lighting? Anybody with experience using color balance filters on still digital cameras?

I figure that since we don't have RAW capture, the more we can do before the sensor, the better.

Every 5D2 video taken under low light has horrible skin tone.. That nasty posterized orange... What is the formula to achieve consistently good/decent looking skin tone? I'm guessing the reason why it looks so bad is because the camera is boosting sensitivity due to low light and red being the most sensitive color channel?

I guess a good question would be does the 5D MKII produce that sickly orange light mainly under wimpy tungsten fixtures? Most household tungsten fixtures are color balanced with a very low CRI and are in the 2600 to 2800 K range, which is very orange.

I would be curious how the 5D MKII reproduces skin tones under better CRI and higher color temp lights at a low output level.

Dan

I guess a good question would be does the 5D MKII produce that sickly orange light mainly under wimpy tungsten fixtures? Most household tungsten fixtures are color balanced with a very low CRI and are in the 2600 to 2800 K range, which is very orange.

I would be curious how the 5D MKII reproduces skin tones under better CRI and higher color temp lights at a low output level.

Dan

You're right, it is mostly under Tungsten...

I know some of you are, so this comment isn't for you, but if you're not manually setting Kelvin or custom WB'ing in low artificial light with the 5D2, your color WILL suck beyond comprehension. The auto WB on the 5D2 in tungsten light sucks so bad it actually is surprising -- at least in my experience. If you search Vimeo for videos, it's plain as day who is running auto and who isn't.

That said, the broader topic of the blue filter in an interesting one. It's worth mentioning that the 5D2 has digital filters available too, but not sure if a real analog filter would probably be better.

WHat do you mean digital filters? The different colored filters are only available for B&W picture styles to achieve different B&W looks.. Unless, I'm wrong here..

I find the default picture styles boost the red chroma way too much. I find using custom picture styles allows you to dial down red chroma to avoid that clipping that looks like orange people. I posted a picture style a while ago that desaturates reds a bit. I'll try to post pictures later.

WHat do you mean digital filters? The different colored filters are only available for B&W picture styles to achieve different B&W looks.. Unless, I'm wrong here..

Yeah, sorry, I'm confusing a few things here. What I meant to reference was "White Balance Correction" (see manual). The manual states, "This adjustment will have the same effect as using a commercially available color temperature conversion filter." I'm not sure if it -ACTUALLY- has the same affect though, maybe an analog filter would still be better. Just saying it's there and it's an option.

I find the default picture styles boost the red chroma way too much. I find using custom picture styles allows you to dial down red chroma to avoid that clipping that looks like orange people. I posted a picture style a while ago that desaturates reds a bit. I'll try to post pictures later.

This is an interesting idea. I've been able to essentially get Custom WB to fix that problem, but as someone who shoots a lot of low light spot-lit music gigs, what intrigues me about this idea is that I often find myself exposing shots to avoid significant red clipping. This causes my overall exposure to be pretty low, causing the need to boost certain levels in post. It's a trade-off I know, and sometimes I'll let red clip just a little so I can bring everything else up at capture, but the point is that I find the reds being the primary driver of my digital exposure in these situations. Using this technique might be able to help me avoid red clipping in a way that lets me boost the overall exposure. Hmmm... Going to have to experiment with that one.

Last week I shot some stuff in the Washington DC subway system, and even though I custom color balanced, the result is kind of a sickly yellow green. I'm thinking that there's just not enough blue (or magenta) light to produce a pleasing image.


The most common problem is the shutter speed being out of sync with the lighting. If the light pulses every 1/60, and the shutter speed is 1/50, the white balance will many times be off. I take a custom white balance at 1/10 to make sure that the pulses are averaged out, but it's still not going to be perfect unless you can exactly match the output of the lights.

Furthermore, when the lights have a low CRI (color rendition index), such as many common fluorescent lights, even the best white balance will result in colors that are off by quite a bit. To resolve this for still photography, shoot a raw still in the bad light with a calibrated color chart such as the ColorChecker, convert that to a DNG file, then pull it into the DNG Profile Editor, and make a profile from the ColorChecker, then apply that profile to the raw files in post processing for excellent color.

I understand that similar ColorChecker-based color correction modules are available for video post processing, but I haven't used them. It would be nice to be able to create a Canon Picture Profile from a ColorChecker shot in the same way as the DNG Profile Editor and just use that before we record video.


I wonder what the optimum color balance is for the 5D?


I haven't seen any spectral sensitivity charts yet.


Maybe, like with RED ONE, it's best to use a blue/magenta filter, color balance with that, and bump the exposure by a couple stops.


It certainly is. Pull a raw file into Rawnalyze if you want to see just how underexposed your blue is. Take a look at a daylight shot too and see how underexposed red and blue are relative to green.


Anybody with experience using color balance filters on still digital cameras?


Yes. I do it to reduce noise, increase dynamic range, and prevent blown highlights.

One of the advantages of a blue filter, as Jim pointed out, is that you can increase exposure and get less noise. But on the 5D2 there is another advantage: less clipping. Even if you don't increase exposure, the blue filter can reduce clipping of the red and green channels. Unlike the RED ONE, the Canon loses one stop of highlight headroom for every increase in ISO. ISO 1600 has 3 stops less highlight headroom than ISO 200. I find that it's a highly useful benefit even when I cannot increase exposure. (And most of the time, when I'm in tungsten, I'm already at the limit of exposure).

A magenta filter is great in almost *all* lighting circumstances, as the green is usually higher than red and blue. Check out Using Magenta Filter for Shooting With a dSLR Camera Under the Daylight (http://www.libraw.org/articles/magenta-filters-on-digicam.html) for more info.

There are several downsides. Color filters do not tend to be the highest quality. Any filter increases flare, but color filters tend to have worse coatings (if any).

It's possible that the color filter will affect colors in ways that are not desirable or expected. Magenta, for example, can affect oranges aquas because the filter spectral response differs from the sensor CFA, so that where it crosses over from its red to green plateaus and green to blue plateaus often misses the wavelengths where the camera's CFA filters are crossing over. If there were lens filters that precisely matched the spectral response of the sensor's color filters, this wouldn't be an issue. I haven't noticed this issue in my own photographs, so I continue to use them.

In landscape photography, or any scene with a lot of shade, the red channel can be heavily underexposed. A red color filter can help greatly there.

I highly recommend having a full set of gels of various strengths, at least blue, magenta, and red. They're inexpensive, and once you figure out which ones and strengths you use the most you can buy those in the form of a glass filter.

Every 5D2 video taken under low light has horrible skin tone.. That nasty posterized orange... What is the formula to achieve consistently good/decent looking skin tone?


Many 5D2 problems combine to result in the "perfect storm" of skin tone issues on the 5D2.

Skipping 2 out of 3 lines for video causes a lot of aliasing artifacts. There is some sort of median filter that is smoothing out the resolution at the highest spatial frequency, and my guess is this is an attempt to slightly reduce the amount of moire. (Yes, it would be even *worse* if Canon allowed the full 1920x1080 resolution to be recorded).

Furthermore, the aliasing artifacts are still so strong that the compression encoder, even at such high bitrates, cannot encode it all, so a lot of details are lost to compression.

For those two factors, using an optical filter to remove resolution will help.

Those two factors alone are enough to result in nasty posterization of the skin, but there's more. The default tone curves do not have nearly enough highlight headroom for many tungsten lighting situations that I find myself in, so the highlights in the red channel (and that channel only) blow, leaving only the green and blue pixels to render detail in the skin, which makes it look plastic.

This can be helped by enabling HTP, which adds another stop of highlight headroom through nonlinear compression, at the cost of more read noise. (Unless you're at ISO 3200, in which case HTP can be enabled for free.) But often the 1 stop offered by HTP is not sufficient, so only a custom tone curve will do, but I haven't yet been able to build a tone curve in Canon Profile Editor that I like that gives me any more than a single stop greater highlight headroom.

The last factor is just the many things that can affect color. White balance, picture profile, low CRI, etc.

Awesome information, Daniel,

My problem shot was with video on a long escalator in the Rosslyn (Arlington) Metro station. One of the problems is that I color balanced at the foot of the escalator, and I think they used different lights in the escalator tunnel. It's possible that the background was also colored strangely. It all combines for a really sickening look. Hopefully, I can get something more pleasing in post.

One problem is that our eyes/brain are so darn good at correcting this stuff. I had no idea that the result would be so bad, based on what I saw - especially since I took the time to make a custom color balance.

After some shooting and testing and more shooting and more testing, this is my re-re-revised favorite picture style. It has nice contrast. It preserves shadow details quite nice. It avoids solarization due to oversaturation without being washed out. It has sharpness dialed all the way down to avoid haloing. Colors also look close to real life when a color temperature preset such as sunlight or tungsten is used. Try it, modify it, comment on it, upload it back modified... Your call. :)

Unlike the RED ONE, the Canon loses one stop of highlight headroom for every increase in ISO. ISO 1600 has 3 stops less highlight headroom than ISO 200. I find that it's a highly useful benefit even when I cannot increase exposure. (And most of the time, when I'm in tungsten, I'm already at the limit of exposure).


Isn't it the other way around? Red one uses software gain only for its ISO adjustments, whereas the 5DMK2 uses hardware gain. Red one records the same data no matter what you set ISO at, whereas hardware gain in the 5DMK2 allows actual increased sensitivity with only minor increases in noise. So might it be more accurate to say that for every stop of iso increase, highlight headroom stays the same but some amount of shadow detail (less than a stop) is lost to noise?

http://www.dvinfo.net/conf/canon-eos-5d-mk-ii-hd/234394-optimum-color-temp.html#post1139918

Red one uses software gain only for its ISO adjustments, whereas the 5DMK2 uses hardware gain.

Correct. RED ONE doesn't allow gain to be changed by any means, whether analog or digital. It's always recorded as metadata only.

The 5D2, on the other hand, uses analog gain up to ISO 3200 and digital gain for 6400+. Canon does not allow any use of metadata gain, and even their digital gains are done in-camera, by deleting information.


Red one records the same data no matter what you set ISO at,


Correct.


whereas hardware gain in the 5DMK2 allows actual increased sensitivity with only minor increases in noise.


That is a common misconception. Sensitivity never changes, on any camera, no matter what the analog gain or ISO setting. On the 5D2, read noise is high at base amplification (23.5 electrons at ISO 100). When ISO is increased, read noise goes *down*. It is lowest at ISO 1600 (2.5 electrons). The RED, on the other hand, has the same read noise at low gain as it does at high gain. They tested it. And since it's the same, they decided to disallow customers from changing it, since the only effect high gain would have is to lose highlight headroom.

Remember that gain/ISO does not cause noise. *Exposure* causes noise. If you reduce exposure, you get more noise. If you increase exposure, you get less noise. The gain/ISO setting is completely separate.


So might it be more accurate to say that for every stop of iso increase, highlight headroom stays the same but some amount of shadow detail (less than a stop) is lost to noise?

No. Here is the longer description of the possibilities:


5D2: For every stop of underexposure (e.g. f/4 to f/5.6), highlight headroom increases one stop, noise gets worse in all zones, and shadow detail decreases one stop.
5D2: For every stop of ISO setting (e.g. from ISO 200 to 400), highlight headroom decreases one stop, noise stays the same in all zones except read-noise dominated zones, such as shadows, where detail improves by some amount (less than a stop).
5D2: If exposure is reduced by one stop and ISO increased by one stop to compensate (e.g. f/4 to f/5.6 *and* ISO 200 to ISO 400), then highlight headroom stays the same, noise gets worse in all zones, but read-noise dominated zones, such as shadows, don't get that much worse (less than a stop worse).
RED ONE: For every stop of underexposure, highlight headroom increases one stop, noise gets worse in all zones, and shadow detail decreases one stop. That's it.

Since the 5D2 has the lowest read noise at ISO 1600, it would be ideal if we could *always* shoot the camera at ISO 1600. But we can't because it has four stops less highlight headroom than ISO 100. Canon has not yet figured out how to get ISO 100 to have the same excellent performance as ISO 1600. Sony and Nikon figured it out, with their A900 and D3X, so hopefully Canon will too, some day.

So obviously it's much better to shoot ISO 100 if you can. It's only when you are willing to sacrifice highlight headroom for improved read noise that it makes sense to shoot higher ISO.

The ideal technology is what RED uses: the read noise stays low no matter *what* the gain is set to. Then we don't have to make the compromise.

All that is to bring me to this point: RED ONE exposed for ISO 1600 has *tons* of highlight headroom (over 6 stops), whereas the 5D2 exposed for ISO 1600 (and set to ISO 1600) only has the normal amount of headroom (~3 stops). This is often not enough for tungsten, so it can be highly beneficial to add a blue filter to the 5D2.

I hope that helps.

Daniel:
It seems like maybe we are talking about some of these issues in different terms.
When I said highlight headroom stays the same as iso increases I meant relative to middle grey as rendered at that iso (perhaps I should have been more specific) but i wasnt assuming just overexposing the image as iso increased, i meant assuming exposure was compensated. In that case, can you explain further how your explanation of what happens in a situation where iso is increased and exposure is compensated in terms of dynamic range and noise relative to the original situation (lower iso and more light)?

Also, I am not following what you mean when you say sensitivity does not change when you increase hardware gain. If increasing iso from 100 to 3200 allows you to under expose by 5 stops and still get good exposure and not lose 5 stops of dynamic range, then how is that not increased sensitivity? Are you talking in terms of a photosite's reaction to a photon or something? Perhaps we are defining sensitivity differently?

Where did you get the information regarding RED's read noise not changing at other hardware gain levels? How is that possible? Doesnt every digital camera decrease read noise with hardware gain increases? Isnt it a shortcoming not to? Given a circumstance where one is shooting in an environment where correct exposure occurs at iso1600, on the RED ONE you are stuck with a noisy image where most of your range is unused (maybe 9bits out of 12) and with the 5DMK2 you can achieve correct exposure (and use all 14bits in still raw mode) using hardware gain and lose much less shadow detail to noise, no? I recall seening video shot at iso 1600 on RED and the same scene on 5dMK2 at iso 1600 and the difference seems massively in favor of the hardware gain method of the 5DMK2 since the RED ONE seems much much noisier at iso 1600 than the 5DMK2. Or is it just the nature of the Canon's class of sensor that is responsible for the difference in performance rather than the hardware gain method?

It seems like maybe we are talking about some of these issues in different terms.


Yes.


I wasn't assuming just overexposing the image as iso increased, I meant assuming exposure was compensated.


I see.


In that case, can you explain further how your explanation of what happens in a situation where iso is increased and exposure is compensated in terms of dynamic range and noise relative to the original situation (lower iso and more light)?


Sure. The effect is different for the RED ONE (metadata ISO) than it is for 5D2 (in-camera ISO):


5D2: If exposure is reduced by one stop and ISO increased by one stop to compensate (e.g. f/4 to f/5.6 *and* ISO 200 to ISO 400), then highlight headroom stays the same, noise gets worse in all zones (due to photon shot noise), but read-noise dominated zones, such as shadows, don't get that much worse (less than a stop worse). Brightness of the file, the image on the LCD, and post-production image stays the same.
RED ONE: If exposure is reduced by one stop and ISO increased by one stop to compensate, highlight headroom increases one stop, noise gets worse in all zones (due to photon shot noise), and shadow detail decreases one stop (less signal no change in read noise). Brightness of the file goes down. image on the LCD stays the same. Post production image brightness stays the same.



Also, I am not following what you mean when you say sensitivity does not change when you increase hardware gain. If increasing iso from 100 to 3200 allows you to under expose by 5 stops and still get good exposure and not lose 5 stops of dynamic range, then how is that not increased sensitivity? Are you talking in terms of a photosite's reaction to a photon or something? Perhaps we are defining sensitivity differently?


I think it's a terms/definition problem as you guessed. Here are what the terms mean to me:

Exposure: intensity of light falling on the sensor.

This is a function of scene luminance, f-number, and shutter speed. Also filters, lens transmissivity, and perhaps other factors. I think this is the original and correct meaning of the word exposure.

Luminance: This is the brightness of the image. It is separate from exposure.

Sensitivity: Response to light.

This is how the sensor responds to light. If it requires 9 photons to cause 1 level change, that is lower sensitivity than a sensor that only requires 3 photons.

Read noise: All noises aside from photon shot noise.

This includes all temporal noises such as reset noise, thermal noise, flicker noise, RTS noise, and amp noise. At base ISO, this type of noise only affects the extreme shadows. The bright parts of the image are only affected by photon shot noise. The higher the ISO, the more exposure zones that it affects.

Here are some examples how these definitions affect the discussion:

In the RED ONE, when the ISO setting is changed from 320 to 640, it doesn't actually change the exposure, sensitivity, or read noise of the camera, it just instructs the post processing to increase the brightness and it also increases the brightness of the display monitor.

In the 5D2, when the ISO setting is changed from 200 to 400, it doesn't actually change the exposure or sensitivity of the camera, but it does decrease the read noise, and it does increase the brightness of the file, and it does increase the brightness of the display.

In the 5D2, when the ISO setting is changed from 200 to 400 and the exposure is decreased one stop (e.g. f/4 to f/5.6), it does change the exposure, does not change the sensitivity, it does decrease the read noise. The brightness of the file and display remains the same. Read-noise dominated zones, such as shadows, don't get a full stop worse.


Where did you get the information regarding RED's read noise not changing at other hardware gain levels?


I thought I read a post by Deanan DaSilva or Graeme Nattress where they tested the RED ONE with higher analog gain and found that the read noise was the same; however, I can't find it now, so I retract the statement.


Doesn't every digital camera decrease read noise with hardware gain increases?


No. It only seems to apply to certain pixel designs. Other pixel designs have the best of both worlds: low read noise *without* increasing gain. The Nikon D3X has a read noise of only 6 electrons at ISO 100, whereas the 5D2 has 23.5 electrons. The 5D2 has to sacrifice 2 stops of highlight headroom by setting ISO 400 in order to get read noise down to 6 electrons. Of course, the 5D2 can get all the way down to 2.5 electrons read noise at ISO 1600, where it can easily beat the D3X, but it would be much better if the 5D2 had 2.5 electrons read noise at ISO 100.

Even Canon cameras only improve up to ISO 1600. 3200 is hardware analog gain, but has no improvement in read noise.


Isnt it a shortcoming not to?


It's a design compromise. The 5D2 has a nice low read noise at high ISO, but at low ISO it's terrible. I'm sure they will figure it out in the future.


Given a circumstance where one is shooting in an environment where correct exposure occurs at iso1600, on the RED ONE you are stuck with a noisy image where most of your range is unused (maybe 9bits out of 12) and with the 5DMK2 you can achieve correct exposure (and use all 14bits in still raw mode) using hardware gain and lose much less shadow detail to noise, no?


First, the exposure (aperture & shutter) is the same on both cameras, so neither is "incorrect", per se. Second, the brightness is the same on both cameras as well (in the case of the RED ONE, the brightness is added to the display monitor and in post production instead of the in-camera file).

Third, the range of bits "used" is the same on both. As explained in the Noise, Dynamic Range, and Bit Depth (http://theory.uchicago.edu/~ejm/pix/20d/tests/noise/noise-p3.html) section of Emil Martinec's essay, none of Canon's cameras suffer quantization error under any circumstances, even when pushing ISO 100 ten stops, so there is literally no benefit to the 14 bits at ISO 1600. In fact, at ISO 1600 you can replace the raw 14-bit Canon data with 10-bit data and lose absolutely no image information whatsoever. Take a look at this visual demonstration:

Canon 40D raw file compared from 14 bits to 7 bits (http://forums.dpreview.com/forums/read.asp?forum=1018&message=31239793)

So essentially your comparison comes down to the fact that at ISO 1600, the 5D2 has less read noise than the RED ONE.


I recall seening video shot at iso 1600 on RED and the same scene on 5dMK2 at iso 1600 and the difference seems massively in favor of the hardware gain method of the 5DMK2


There are other cameras with hardware gain, such as the D2X, that can't hold a candle to the RED ONE. What matters is the read noise, and Canon is the king of random read noise.

I hope that helps clarify things.

After some shooting and testing and more shooting and more testing, this is my re-re-revised favorite picture style. It has nice contrast. It preserves shadow details quite nice. It avoids solarization due to oversaturation without being washed out. It has sharpness dialed all the way down to avoid haloing. Colors also look close to real life when a color temperature preset such as sunlight or tungsten is used. Try it, modify it, comment on it, upload it back modified... Your call. :)

Thanks Luis, going to have a play with that later. Looks, on the surface of it, quite natural and a little filmic. Migh be good for a little shoot I have planned for next week.

Cheers

Avey