RAW sensor response and strobe color temps

Long ago I wondered what the native color response of dSLR sensors was. Such data isn't readily available but there must be plenty of experts who know the answer since writing a raw converter would require such knowledge.

Recently while researching color calibration scripts for ACR, I stumbled upon a thread where photographers had devised a means of testing this very thing and were advocating the use of filters to optimize dynamic range. This is exactly what I had hoped for and wanted to do. I turns out that dcraw, an open source raw converter than many commercial products haved been based on, offers the ability to disable white balance processing. The program is a command line utility and its options change over time, but with a little experimenting I got what I wanted.

The trick is to disable all color processing and gamma correction and simply demosaic the image and produce a TIFF file. The result is a linear-encoded file without any gain added so it looks quite black. In photoshop you can assign a linear profile to help make it look better but it's really irrelevant how it looks. What you want to do is shoot a calibrated gray reference, scale it reasonably, select a small region, and measure the pixel values between R, G, and B. A little computation later you have the exposure information you want.

I've done that with the two bodies I own, a D2xs and a D80, and I've tested it with a single Z240, the only strobe I've received so far. I tested the Z240 with and without the 0.5 diffuser and there's only a mild effect. Here are the results:

D2xs: R: 0ev G: +0.80ev B: +0.35ev

D80 : R: 0ev G: +1.05ev B: +0.95ev

I shot and converted multiple times to be certain the results are repeatable but there is some room for error. I'd take these as good estimates but not accurate to two significant digits. ;-) At first I'd planned to show what these colors looked like, but I realized without having applied the color profile of the device such a sample would be meaningless. Applying such a profile would result in a color we all know...gray! Needless to say, the images look pretty green.

Anyway, what this says to us is quite interesting. The green channel is the most sensitive and I believe that is true for most Bayer sensors. The D80, when used with a Z240, gives up and entire stop of dynamic range in the red channel. In other words, with that strobe the camera may well be ISO 100, but the red channel is really ISO 50. Green and blue are quite well matched in the D80 so only the red channel suffers. With this setup, if a user were to shoot with a CC30R filter he would maximize the performance of his system at the cost of one stop of strobe power. I wouldn't even mind putting that filter on the strobe itself.

The D2xs is not quite so straightforward. For macro, the strobes could be filtered so long as you only wanted black backgrounds. Otherwise, you will probably want to filter the lens and combinations of filters can be hard to use without optical degradation. The D2xs with the Z240 needs about 15CC of magenta and 10CC of red filtration to be optimized. Depending on the type of shooting, different combinations may make sense. Warming the Z240 the right amount would leave only a magenta filter on the lens, probably a CC20M. That is what I intend to pursue.

Each camera model should be tested. You can't assume what a D200 will be like from what a D80 is like. You don't know what a Canon will do by either of these tests. I believe it is a common trend, though, that the green channel is the most sensitive.

I think this exercise raises some points. First, warm strobes are good for digital, even macro. Traditional wide angle strobes may be warmer than we need, though. The Z240 is slightly too cold for either of these cameras and water makes it worse. Second, we may be giving away more potential performance than we realize. It's well known that raw converters have some highlight recovery headroom and some of that may be the result of this imbalance. Without much trouble, we could gain 1/2 of dynamic range with the D2xs and a full stop with the D80 when using Z240 strobes. Considering how hard the manufacturer works for such a gain I'd like to get that back!

I intend to run full color calibrations with any such filters but I cannot until I have my complete rig. Shooting a ColorChecker chart requires very careful illumination and I can't do that yet with strobes. I would like to demonstrate that such unconventional techniques will not result in odd color rendering (or at least not waste any underwater time trying them!). Stay tuned...

Hi Craig,

thank you. This is great information.

The only transmission diagram about Bayer filter array transmission I’ve seen so far is from this site (worth reading): http://www.molecularexpressions.com/primer...agesensors.html

(I think Bart posted this quite a while ago)

I was surprised that there aren't steep bandpass effects with steeper declines. Which means for example red wavelength even stimulate the green channel to some degree.

Bayer sensors have the double amount of green filters compared to red or blue ones. Your results match in terms of bright green exposure.

Shooting with filters, I more and more used combinations with stronger green suppression. I took one shot topside in sunlight with over 200 mireds and over 75 CC M. It was possible to obtain natural colours during RAW conversion for this topside sunlight shot (without major quality loss). I think this means there was still enough green exposure present. More than I expected.

I think it’s easier to recover less green than less red if you take somewhat 5000K and Tint 0 as reference in the RAW converter. Because according to your results, such kind of “neutral†setting is already based on a brighter green channel exposure compared to red.

An exchangeable filter array would be a great accessory for my 5D! A dedicated underwater shooter’s filter array together with a profile for RAW converters would be desirable B)

Julian

OK Craig. For me reading the stuff you write is like reading a textbook on the subject, and I am very happy to come away from it with a basic understanding (I hope).

Practical question: I am headed to Galapagos next Saturday with my D200, WA setup, and Subtronic strobes. I believe these are slightly warmer than Inon's. I know such a set up has not been tested, but would you suggest trying a CC30R filter on my lens and/or strobe. On strobes what type of filter material etc and from where do you purchase?

OK Craig. For me reading the stuff you write is like reading a textbook on the subject, and I am very happy to come away from it with a basic understanding (I hope).

Practical question: I am headed to Galapagos next Saturday with my D200, WA setup, and Subtronic strobes. I believe these are slightly warmer than Inon's. I know such a set up has not been tested, but would you suggest trying a CC30R filter on my lens and/or strobe. On strobes what type of filter material etc and from where do you purchase?

The D200 and D80 share similar but not identical sensors and it's really the Bayer filter array that's being tested. For that reason, the D200 really needs to be tested separately. I'd expect Nikon sensors of similar design and manufacture to behave similarly but you never know.

The only testing I've seen outside mine involved Canon sensors. They had much stronger green sensitivity than blue; much like my D2x.

Warmer strobes will have more red than blue. Think of it, in rough terms, that the green portion is fixed, red grows stronger and blue grows weaker. Those are really rough terms. Your strobes and camera really need to be tested together.

I worked out my D2x/Inon numbers more carefully and I found that an 81B warming filter on the Inons in combination with a CC20M was just about perfect. The warming filter on the Inons would put them around 5000K, pretty close to the warm wide angle strobes available. With warmer Subtronics or Ikelites, A CC20M would be perfect for my camera.

If were to guess, a CC20M on your lens and nothing on the strobes is where I'd start. Lead times on various filters can be pretty long though. I use B&H and orders from them can take several weeks. Testing your camera is quite easy though and it only takes a calibrated gray card. I used a Whibal card myself, but a step chart or ColorChecker works too.

P.S. A couple things. I'd be happy to help anyone who wished to test this with their rig. It's quite easy and I'd be interested in the results. Also, anyone trying this should not be surprised if their image reviews look a little off and their raw converter white balance gets skewed. No cause for concern!

Jolly,

Yes, bayer filters have surprising color overlap and I think they need that to work properly. My understanding is that it is quite expensive to produce really good filters so filter quality does vary somewhat between cameras. It's a bit mysterious. CMOS may work a bit differently than CCD. Don't know.

I agree with you that green seems more easily recovered than red and this may explain why. I've always had good luck putting green back in after removing it with a filter as well. It's good to hear others with similar experiences.

In a conversation with Ryan regarding complimentary magenta/green filters, he said that sometimes he prefers to shoot with a stronger magenta filter than green. At this point it seems clear that he is actually improving the performance of his sensor by doing so. In my case, it seems that a CC50M with a CC30G filter would be the right complimentary combination. Seems like a lot, but we have to remember we're only removing light we don't want! I may consider a 40/20 combination as well. We'll see.

In the old days, there was an UW filter manufacturer named Aquacolor who supplied filters to Amphibico. I actually preferred theirs to UR Pro. They were more purple than orange and were available in 5 foot depth rating increments. The deeper filters were incredibly dark yet cameras worked great with them. No reason why a CC50M can't work just as well.

If were to guess, a CC20M on your lens and nothing on the strobes is where I'd start. Lead times on various filters can be pretty long though. I use B&H and orders from them can take several weeks. Testing your camera is quite easy though and it only takes a calibrated gray card. I used a Whibal card myself, but a step chart or ColorChecker works too.

 

P.S. A couple things. I'd be happy to help anyone who wished to test this with their rig. It's quite easy and I'd be interested in the results. Also, anyone trying this should not be surprised if their image reviews look a little off and their raw converter white balance gets skewed. No cause for concern!

All I can seem to find for my 17-55 at such short notice is aTiffen CC30M. I ordered it and will give it a try next week. When I get back I'll take you up on your offer and bug you as I try to understand this stuff better.

Craig,

I have just used a complimentary filter on strobes which “just†counteracts to the lens’ filter. When I set WB and Tint to neutral settings, the strobed colours are quite right. But of course, this does not take into account what you have pointed out. I just used neutral ACR settings as reference.

I just downloaded DCRAW. What options did you add in the command line? Is ‘-M†(don’t use an embedded colour matrix) one of them?

Thank You.

Julian

Julian,

Very timely! I have an article in review with some comparison photos using complementary filters vs no filters. Hope to publish it soon!

Cheers

James

Julian,

I'm glad that worked for you. That's how it's supposed to work. ;-) Hopefully you liked the results.

dcraw options are all over the map it seems. I'm using dcrawx86 version 8.72 for OS X, and my options are very different from even the PowerPC versions that came in the same bundle!

The key option for me is -D. That option performs a demosaic but disables all other processing. There is a -d option that even disables demosaic but then you can't tell the colors apart :-( These options together are called "document mode" whatever that means.

I also use -T for TIFF output, -h to reduce size, and -4 for 16 bit linear output.

You may need to study your version. Other versions use the option to specify all the gain coefficients. If that's your version, set them all to 1.

After I get a conversion, I load the TIFF into PS and run an autolevel. You can immediately see if there's clipping using this technique. I've been doing this to see if there's actual clipping because both the camera and the RAW converter will lie about it!

Finally, what I do is select a white patch, then use the histogram to select a channel and read the mean. You can then compute the Ev offsets by taking the base 2 logs of the ratios of the means.

craig

I've been playing some with this and gotten some surprising results. The color temp of the Inon strobe is a bit cold so applying a warming filter has had the expected positive results. I'm finding an 81B to work well.

I initially felt a CC20M filter would be a good match but have subsequently found that 40M is actually required. That came as quite a surprise. Also, because the magenta filters have a small effect on blue relative to red, I need to back off on the warming filters on the strobes.

The problem here is that the bayer filters have really broad response so and magenta CC filter is causing some collateral damage to red and blue.

I stumbled across this webpage: Nikon RAW gain coefficients

Basically, it's a big table of several Nikon DSLRs and the RAW gain coefficients used under a variety of lighting. Ideally, we want the coefficients to be (1,1,1). You can see in the table that different cameras produce surprisingly different results.

I discovered two things today. The first is that the tint control in ACR has units that roughly correspond to CC units of magenta/green filters. I did not know that! The second is that the RGB channels clip at different values close to 4095. Because of that, the ideal gain values aren't exactly (1,1,1) and you need to expose a step chart to know for sure. I'm disappointed by that.

Now I'm going to violate common sense and link to this Luminous Landscape forum topic for two reasons. First, they are discussing this very topic, and second they are advocating CC40M filters and that matches my recent observations.

Just providing an update.

I've tested various warming filters and magenta filters in hopes of finding the perfect match between my Inon Z240 and D2xs. The warming filters worked exactly as I expected but the magenta filters had a far greater negative effect on reds than desirable. This is due to their considerable effect on yellow light. The optimal combination I found was an 81A in combination with a CC40M. I also tried flourescent Day filters since they are these types of combinations. The Hoya FL-D was a good but not perfect match.

Subsequently, I discovered that the camera saturates at slightly different values for red, green, and blue. Specifically, the 8 bit values for my camera are 254, 240, and 251. As a result, my initial results were off. The more accurate numbers are (0. +0.73, +0.31). The old results were (0, +0.81, +0.33). I have slightly less to correct than I thought.

I decided to try the Rosco pink filters since they are essentially magenta filters with yellows not blocked. The results of the CC15P were (0, -0.42, -0.30) and the CC30P was (0, -0.83, -0.60). Neither of these is a perfect match; each yields an imbalance of about .3 stops. One saturates green first, the other saturates red.

Considering that either of the Rosco pink filters results in a mismatch of about .3ev, the CC15P looks like a good choice. It is rated at -0.5 stops but yields only 0.4 stops of extra red dynamic range with my strobes. With a camera with more blue sensitivity, the CC30P would be better.

I have run full camera calibrations with outstanding results on one of my filter combinations but none yet with the pink filters. When I do so I will post those results. One of the issues with this filter source is that they are only available in larger, polyester sheets (for lighting) so you must be careful that they don't degrade image quality. I plan to run some IQ tests using Imatest to make sure images aren't softened.

The beauty of a CC15P is that it does no damage to red light ISO and reduces ISO overall much less than a CC40M. Hopefully they will work well. Rosco Pink filters look suspiciously like the old Aquacolor video filters!

For those not familiar with Aquacolor, look here: http://www.aquacolor.com/Smptestrt.htm

That article was presented in 1983!

After researching the removal of AA filters, I stumbled upon the explanation for why these cameras have native white balance that is so unusual. Check out this conversion:

The 300D filter Modification

Look at the standard filter that he has removed and you will see that it is not neutral in color but cyan. The IR-UV cut filters used in these cameras are tinted cyan!

I contacted maxmax.com, a company that does IR, UV, and HR conversions and asked them about it. They confirmed that most all cameras use this type of colored filter. Unfortunately, the filter they use for IR-UV blocking is also cyan and is even darker. This is not the case for all blocking filters but it is the case for theirs. So far, they haven't expressed interest in offering an alternative filter though they have suggested using clear glass and installing the filter externally. Obviously, external filters are problematic for some of the lenses we use underwater.

It would seem that removal of the AA filter and the UV-IR cut filter would be fine for UW shooting so long as your strobes are filtered to remove UV output (diffusers should work for that). Unfortunately, split shots would be out unless a filter was added to the lens, and those shots use lenses that are least likely to take filters. :-(

Hi Craig,

Whoa! That is very surprising and must certainly have an effect on underwater images not anticipated by the camera/sensor manufacturers.

We were talking about this and you'd told me you were considering using a CC10m filter in place of the AA filter. Have you pursued that too?

Cheers

James

Yes, I asked about that.

MaxMax, the company that does the conversions, has blocking filters cut down and polished for each camera they support. They haven't been receptive to doing a custom filter for me that is neutral. My first request was for them to use a CC30M. Those are available in crown glass and block both UV and IR effectively. I then asked if they would consider using a traditional blocking filter that isn't colored and sent them a link. He wasn't going for that either. I finally asked if they could rig a gel in addition to simple glass. He said that was a bad idea. He doesn't want to do special work, although he did say he would consider converting a D3.

My main problem with going with straight glass is that fisheyes and the 14-24 don't take gels or filters very readily. Macro is not a problem since I don't need a filter for that and I have room for one anyway. I'd also have to consider how I'd shoot wide angle on land. Basically, I will always need a rather expensive filter on my lens.

BTW, I have my D3 and 14-24 now. The 24-70 is late.