Luminous Landscape Forum
Equipment & Techniques => Digital Cameras & Shooting Techniques => Topic started by: Willowroot on May 17, 2006, 02:17:10 pm
-
I just got a K-M A2 so I've been looking around at digistuff ... ran across this interesting technique at DPReview. Essentially what you do is use a CC040M magenta filter to limit the amount of green light that the sensor sees. White balance with the filter on of course. Since the green channel is the first to blow in daylight, this allows the red and blue channels to "catch up" and the practical effect is to expand the effective dynamic range of the system.
Here's the links:
thread one (http://forums.dpreview.com/forums/read.asp?forum=1021&message=17181241)
thread two (http://forums.dpreview.com/forums/readflat.asp?forum=1021&thread=17490123)
Would be interested in the comments of people here
-
Since the green channel is the first to blow in daylight, this allows the red and blue channels to "catch up" and the practical effect is to expand the effective dynamic range of the system.
It seems that bjanes is a great fan of this technique, but I'm still skeptical. The general principle for use of filters is that you use the filter to suit the scene. In this use of filters which you've referred to, the use is to suit the characteristics of the sensor, completley ignoring differences in scene content. One can't help wondering why Nikon or Canon did not reduce the sensitivity of the 'green' pixels, through one way or another, if this has the effect of increasing dynamic range. For some cameras, what that would mean is that the true ISO was brought back into line. For example, ISO 100 on the D60 is more like ISO 160. Reduce the sensitivity of the 'green' pixels and you are back to a true ISO 100 with (perhaps?) a half stop increase in dynamic range.
My gut feeling is, such an approach (using a filter to suit the sensor rather than the scene) will result in a botched effect, the benfits of which will vary according to scene content and color values.
I've brought up before (in the 'interesting article' relating to ETTR), the issue of 'true blue' skies shifting towards cyan during overexposure. I'm going to post an example of this effect and then try to predict what might happen if I had used a magenta filter. I'm really thinking while I'm writing because I don't have any hard evidence.
The photo, by the way, is a crop of the top of a vertical shot. This is not overexposed as a result of not knowing what I was doing, but deliberately overexposed because it was a snapshot of my partner who was the focus of interest. I'm not sure why so many people have an irresistable urge to place themselves in front of all interesting, beautiful, old and valuable monuments, buildings, artifacts and simple beautiful scenes, when photos are taken. Do they feel a need to compete for the attention of the photographer, perhaps? Or do they simply want a record of the fact that they were there, in case anyone might disbelieve them.
The shot has a -2EC adjustment in ACR.
[attachment=571:attachment]
The RGB values for the 'true blue' part towards the upper left corner are R=104, G=114, B=143. There's a clear predominance of blue, as one would expect.
The RGB values within the part encircled by the red line are, R=208, G=219, B=228.
In other words, green has become more prominent as a result of the blue channel clipping. (Pure cyan is a mixture of blue and green in equal parts).
How would a magenta filter affect this? (Magenta being equal parts of blue and red).
Well, I'm relying upon logic here because I don't have a magenta filter. It seems to me, in this example, the situation would be complicated. The magenta filter would suppress the green channel which is not a problem. The problem is, the blue channel is clipping. If the exposure is the same, the blue channel will still clip despite the magenta filter. However, the green channel will be subdued so that the area enclosed by the red line in my example should (might?) begin to look more 'blue' than 'cyan'.
Is that a solution to the problem? If the exposure is the same, I've done nothing for the shadows. It's true, I've corrected a cyan shift in one part of the sky whilst simulataneously creating a magenta shift over the rest of the image.
If I can correct that magenta shift in the rest of the image successfully, then perhaps I could argue that I've been able to use a 1/2 stop or 2/3rds stop greater exposure, which would help the shadows. On the other hand, if I know what I'm doing in Photoshop, I can isolate that part of the sky which suffers a cyan shift, and change the hue in 'Hue/Sat' control.
Where's the benefit?
I should point out, in case anyone is wondering where the cyan shift is, the sRGB jpeg compression reduces this difference. I work in the ProPhoto RGB space and convert to sRGB for net display.
-
I just got a K-M A2 so I've been looking around at digistuff ... ran across this interesting technique at DPReview. Essentially what you do is use a CC040M magenta filter to limit the amount of green light that the sensor sees. White balance with the filter on of course. Since the green channel is the first to blow in daylight, this allows the red and blue channels to "catch up" and the practical effect is to expand the effective dynamic range of the system.
[a href=\"index.php?act=findpost&pid=65806\"][{POST_SNAPBACK}][/a]
What is white balance application? Basically just a shift of the values of one channel relative to the others.
If I am not mistaken, what you do here is under-expose the green channel at capture, but the white balance will basically then bring back that channel to where it belongs so as to get neutral colors. Since the white balance is just one point along the curve, I am not even convinced that you will completely get rid of the green cast with white balance click only, but let's assume that you do.
You might be gaining something in the highlights (depending on how the white balance exactly works with the channels), but you are also very likely to get a more noisy green channel, since you are doing the very opposite of the recomended expose to the right dogma.
Anyway, it might work in some cases and for some cameras whose green channel isn't too noisy in the shadows.
Cheers,
Bernard
-
You might be gaining something in the highlights (depending on how the white balance exactly works with the channels), but you are also very likely to get a more noisy green channel, since you are doing the very opposite of the recomended expose to the right dogma.
[a href=\"index.php?act=findpost&pid=65867\"][{POST_SNAPBACK}][/a]
Exactly! In my example above, comparing a cyan shift corrected in PS as opposed to a cyan shift corrected by use of a magenta filter, one has 2 shots with the same 'external' exposure. The shadows, however, having less 'internal' exposure in the green channel, with use of the magenta filter, will actually be noisier to some degree.
On the other hand, blown detail in the clouds will be brought back slightly. Net effect regarding dynamic range... a big fat zero.
-
see following post
-
What is white balance application? Basically just a shift of the values of one channel relative to the others.
If I am not mistaken, what you do here is under-expose the green channel at capture, but the white balance will basically then bring back that channel to where it belongs so as to get neutral colors. Since the white balance is just one point along the curve, I am not even convinced that you will completely get rid of the green cast with white balance click only, but let's assume that you do.
You might be gaining something in the highlights (depending on how the white balance exactly works with the channels), but you are also very likely to get a more noisy green channel, since you are doing the very opposite of the recomended expose to the right dogma.
Anyway, it might work in some cases and for some cameras whose green channel isn't too noisy in the shadows.
Cheers,
Bernard
[{POST_SNAPBACK}][/a] (http://index.php?act=findpost&pid=65867\")
Bernard (and Ray),
To understand the use of the CC 40M filter, first look at the relative sensitivities of the channels exposed under daylight. Here is an example with the Nikon D200. The PS histogram shows the channels of the light gray patch of the Macbeth CC as they are recorded in the raw file and converted without white balance or gamma correction in DCRaw . The image is dark (gamma 1) and the colors are posterized because this is a GIF to show the menu letters without jpeg artifacts:
[attachment=575:attachment]
Here is a similar image for the EOS 1D Mark II in jpg, with exposure such as to give a value of around 240 in the white square of the color checker.
[attachment=576:attachment]
As is evident with both cameras, the green channel shows the higest response followed by blue and then red. To white balance the picture, the raw converter for the Nikon image multiplies the red channel by 1.83 and the blue channel by 1.36, assuming exposure at 5400K as indicated by the chart below:
[a href=\"http://www.pochtar.com/NikonWhiteBalanceCoeffs.htm]http://www.pochtar.com/NikonWhiteBalanceCoeffs.htm[/url]
For the Canon, the multipliers can be calculated from the raw channel data. The red multiplier is 1.92 and the blue 1.29.
Now if you expose fully to right in the green channel without the filter, the red and blue channels are not fully to the right, are effectively underexposed, and will suffer in number of tones and noise. By placing the magenta filter over the lens, you hold back some of the green light to better balance the channels, allowing ETTR with the red and blue channels also. You increase the exposure after placing the filter so that now all three channels are exposed to the right. Since the green channel is fully exposed (not underexposed), it does not suffer any increased noise or other ill effects. For WB you can either preset it with a gray card or include a neutral reference in the picture.
To see the true values in the channels on the histogram, you must view them before the white balance multiplier is applied with DCRaw (after the fact), or you can load a special white balance (UniWB) into the camera that sets the blue and red multipliers to 1.0. The camera histogram will then display the channels correctly and you can use them for ETTR if your camera has color histograms. You use this hack, then you will have to set the WB properly in the raw converter. Unfortunately, this hack confuses Adobe Camera Raw into thinking it is dealing with a double exposure with Nikon cameras, which set the WB coefficients to 1.0 in this situation. With ACR you can still WB with the eyedropper, but presets will not work correctly. Nikon Capture works normally in this situation.
This method allows better ETTR in all channels and is no cure for massive overexposure of the blue channel such as in Ray's example, which I think is irrelevant to the method. If you are exposing the blue more to the right, then there will be less headroom in the blue channel and the clipping that Ray observed would be worse, but as Bruce Fraser has pointed out, ETTR is not overexposure. Ray will have to learn to cope with high dynamic range situations.
Daylight is not the only situation where filters can improve noise and dynamic range. With tungsten (which is quite deficient in blue), the blue channel is quite underexposed and needs a multiplier of 2.29 with 3200K for the D200. In this situation, one can get better results by using an 80A filter just as with film photography. Of course, one loses 2 stops and the preview appears blue.
The CC40M filter reportedly only allows about 2/3 stop more dynamic range and may not be worth it if you are using a camera with very low noise, such as the Canons. However, in situations with very wide dynamic range such as landscapes, it may be worth a try. Personally, I do not have the magenta filter and have not tried the method. However, I think it is a very interesting concept that demonstrates some important facets about white balance.
However, I think that CC40M is too strong. If I were to buy a magenta filter for the Canon, I would get a CC20M. The green density os 0.2 or 63% transmission of green, or -2/3 EV. It would bring down the green channel to 93, placing it midway between the red and blue channels and thereby allow +2/3 EV more exposure to the right without blowing the green channel.
-
Exactly! In my example above, comparing a cyan shift corrected in PS as opposed to a cyan shift corrected by use of a magenta filter, one has 2 shots with the same 'external' exposure. The shadows, however, having less 'internal' exposure in the green channel, with use of the magenta filter, will actually be noisier to some degree.
On the other hand, blown detail in the clouds will be brought back slightly. Net effect regarding dynamic range... a big fat zero.
[a href=\"index.php?act=findpost&pid=65875\"][{POST_SNAPBACK}][/a]
Ray,
You do not understand this method at all. Please read my earlier post before coming to rash and erroneous conclusions and let us know what you think.
-
As is evident with both cameras, the green channel shows the higest response followed by blue and then red. To white balance the picture, the raw converter for the Nikon image multiplies the red channel by 1.83 and the blue channel by 1.36, assuming exposure at 5400K as indicated by the chart below:
Bill,
Would you have any insights about the reason for this response, G>B>R ?
Thanks! Peter
--
-
Bill,
Would you have any insights about the reason for this response, G>B>R ?
Thanks! Peter
--
[a href=\"index.php?act=findpost&pid=65942\"][{POST_SNAPBACK}][/a]
Peter,
That's a good question. On the DPReview Nikon forum it was suggested that the sensor has to perform well under other illuminations. However, for tungsten, more blue sensitivity would be desirable, so if that were the case I would expect more blue response.
If I were designing a sensor, I would try to have the sensitivities equal for daylight, but I am not an engineer and do not really know the answer; however, since both Nikon and Canon have excellent engineers, and there must be an explanation rather than manufacturer stupidity.
Bill
-
Ray,
You do not understand this method at all. Please read my earlier post before coming to rash and erroneous conclusions and let us know what you think.
[a href=\"index.php?act=findpost&pid=65917\"][{POST_SNAPBACK}][/a]
I think I do understand it. Perhaps it's you who doesn't understand the use of filters.
Let me go through it again and if I make an incorrect statement, you can point it out and explain where I'm wrong.
1. The red and blue pixels (photodiodes) appear to be less sensitive and noisier than the green pixels.
2. The green pixels are the first to clip (all else being equal, which it never is).
3. Introducing the right type of magenta filter suppresses the green component of light, and to a lesser extent, the blue channel, which allows for a greater exposure in some situations.
4. The greater exposure, partly offset by an over all reduction of light as a result of the filter, can reduce shadow noise through an increase in the red and blue channels, at the same time avoiding clipping of highlights in the green channel which, without use of a filter, would clip at the same exposure.
5. The net effect, all being well and fingers crossed etc, is a real increase in dynamic range, which as you say, might be as much as 2/3rds of a stop, but I've yet to see an example.
Now for the tricky bit. Any increase in dynamic range is totally dependent upon being able to increase exposure without clipping any of the channels. Increasing exposure is not always possible when using a filter that is not selected specifically for the scene being photographed.
This is the major flaw in the process. I'll give 2 examples which I predict would have a completely opposite effect on DR when using a magenta filter.
1. The best case scenario would be a scene with strong greens and cyans in the bright parts and saturated reds and magenta in the shadows. The suppression of green in the filter would facilitate a substantial increase in exposure without clipping the highlights, and such increased exposure would be 'real' for the red channel in the shadows since the magenta filter does not suppress red. The reds (and to a lesser extent the blues) would therefore be stronger and less noisy.
2. A worst case scenario would be a scene with strong reds in the bright parts of the image (a flaming red sunset, for example) and saturated greens in the shadows (foliage, freshly fertilised lawn, leguminous plants that get their nitrogen from the atmosphere etc). In such a situation, there's no possibility of increasing exposure as a result of the magenta filter, because red is the critical channel that's going to be blown, with or without filter. Use of a magenta filter in these circumstances would have a disastrous effect on the predominantly green shadows because the filter reduces the intensity of green light, the opposite of what you want.
The result in this situation would be less dynamic range.
And of course, I'm sure you'll agree, there'll be a whole range of scenarios in between those 2 extremes. Some might result in a worthwhile increase with use of a filter. Some might result in zero increase in DR and some even a negative increase in DR.
It's ironic that in another thread you've made the point about the difficulty of getting sharp results with a 22mp camera, yet here you are advocating reducing shutter speed by a whole stop or more.
-
Bernard (and Ray),
To understand the use of the CC 40M filter, first look at the relative sensitivities of the channels exposed under daylight. Here is an example with the Nikon D200. The PS histogram shows the channels of the light gray patch of the Macbeth CC as they are recorded in the raw file and converted without white balance or gamma correction in DCRaw . The image is dark (gamma 1) and the colors are posterized because this is a GIF to show the menu letters without jpeg artifacts:
[a href=\"index.php?act=findpost&pid=65916\"][{POST_SNAPBACK}][/a]
Thanks for the feedback. Interesting read.
I might give it a try during a coming shoot.
Cheers,
Bernard
-
Now for the tricky bit. Any increase in dynamic range is totally dependent upon being able to increase exposure without clipping any of the channels. Increasing exposure is not always possible when using a filter that is not selected specifically for the scene being photographed.
This is the major flaw in the process. I'll give 2 examples which I predict would have a completely opposite effect on DR when using a magenta filter.
2. A worst case scenario would be a scene with strong reds in the bright parts of the image (a flaming red sunset, for example) and saturated greens in the shadows (foliage, freshly fertilised lawn, leguminous plants that get their nitrogen from the atmosphere etc). In such a situation, there's no possibility of increasing exposure as a result of the magenta filter, because red is the critical channel that's going to be blown, with or without filter. Use of a magenta filter in these circumstances would have a disastrous effect on the predominantly green shadows because the filter reduces the intensity of green light, the opposite of what you want.
The result in this situation would be less dynamic range.
[a href=\"index.php?act=findpost&pid=65956\"][{POST_SNAPBACK}][/a]
In your second example let us assume that the luminance of the red is the highest in the image and you will expose so as to place the reds at the right of the histogram. I don't know what white balance you plan to use, but I would use daylight so that the warm colors of the sunset would be preserved. If you choose to use WB other than daylight, use of the magenta filter would not be appropriate and this case will not be considered further.
I don't know the multipliers for your 5D, so I will use those for the EOS1D Mark II and assume they are similar. Because the red multiplier for daylight white balance is 1.92, your reds (which are exposed to the right with white balance) are only midrange in the raw file, effectively 1 stop underexposed. They are in no danger of blowing out.
So you add the magenta filter and increase exposure by 2/3 EV. The increased exposure compensates for the reduction of green by the filter, and the amount of light falling on the green sensors is the same. The red channel is now toward the right by 2/3 EV and will have a better signal to noise ratio and dynamic range. The image is improved rather than the disaster you predicted. What you failed to do is to analyze what is going on in the raw channels.
It's ironic that in another thread you've made the point about the difficulty of getting sharp results with a 22mp camera, yet here you are advocating reducing shutter speed by a whole stop or more.
[a href=\"index.php?act=findpost&pid=65956\"][{POST_SNAPBACK}][/a]
I am assuming for a static situation such as a sunset you would be using a tripod
-
I don't know the multipliers for your 5D, so I will use those for the EOS1D Mark II and assume they are similar. Because the red multiplier for daylight white balance is 1.92, your reds (which are exposed to the right with white balance) are only midrange in the raw file, effectively 1 stop underexposed. They are in no danger of blowing out.
If the reds are only midrange without the filter, then I am not exposing to the right. Because it's late in the evening, there may be a great expanse of green in shadow which I don't want to look noisy or totally black. I'm in a high dynamic range situation with the highlights predominantly red. I increase exposure till the red channel on my histogram is pushing against the right (the 5D has an RGB histogram). The green channel is looking pretty weak though and much of it is comprising the yellow of the sunset which is, however, more red than yellow. I sense I'm going to have more noise in the shadows than I'd like. I think of bjane's clever trick of using a magenta filter which I just happen to have in my pocket.
I slip on the magenta filter. The red channel is unaffected because of course it lets through all the red light, but the blue channel is now worse and the green channel is much worse. I can't increase exposure because the red channel is already as far to the right as it can be without clipping. The filter is not only useless in these circumstances, if I were to use it it would actually decrease dynamic range.
Is this not the case?
-
I am assuming for a static situation such as a sunset you would be using a tripod
[a href=\"index.php?act=findpost&pid=65973\"][{POST_SNAPBACK}][/a]
Not necessarily. Sunsets can be very bright when the camera is directed at the sky and an exposure to the right can be quite short. I've taken many hand-held shots of sunsets. I generally use a tripod only when I have to, and that is mainly for stitching purposes.
-
Would you have any insights about the reason for this response, G>B>R ?
Maybe just the inks? Some years ago, Kodak (and Nikon) experimented with CMY CFA'a for this reason. Does anybody know, why this technique died? Any major flaws?
If the reds are only midrange without the filter,
Ray, how do you meter a scene, if you want to measure only the light, and want to bann all reflective issues and influences of colors to the meter? If you would use a grey card, then the magenta filter method would be fine with you.
Colors of the subject play no role at all. It's all about adopting different sensitivities of the channels of a certain sensor to a certain light (temperature). A very easy example to understand is the situation with a tungsten lighted scene. The uncorrected image will have a orange color cast, since the spectrum of the light has much red and little blue. So the blue channel will be exposed purely, the green a bit more and the red a lot more. If you want to handle this problem with software based white balance only, you have to increase the values of the blue channel exessively and the green ones a bit less to match the red channel values in order to get a neutral grey, right? So you will introduce much noise in the final image. Therefore, it would be a way better idea, to balance the light before it reaches the sensor, eg using a blue 80B filter. This filter will block a certain amount of the red light, so that (ideally) the light after passing the filter has about equal parts of blue, green and red. Of course, you now have to expose longer or open up the aperture (or raise the ISO), and that might be unwanted or impossible in some situations. But the light is better balanced, and the image will suffer less noise. So what you do here, is to balance the incoming light. The colors of the scene are completely irrelevant.
The above proposed method is all the same, it balances light to the sensors needs. This balancing has nothing to do with colors in the scene, like green foliage or red skyes. It only has to do with the light itself and the sensor. So you'd need a different setup for each sensor and light temperature. If that's worth the hassle, is an individual question.
The other question is, if you want the light to be balanced. A beautifill colored sunset will loose everything, if you match the WB exactly to the temperature of light. So sometimes a color cast is exactly what you want.
But as a matter of principle, the filter method is absolutely correct. It's always a better idea, to balance the light before it hits the sensor, rather than balancing it afterwords by software.
-
If the reds are only midrange without the filter, then I am not exposing to the right. Because it's late in the evening, there may be a great expanse of green in shadow which I don't want to look noisy or totally black. I'm in a high dynamic range situation with the highlights predominantly red. I increase exposure till the red channel on my histogram is pushing against the right (the 5D has an RGB histogram). The green channel is looking pretty weak though and much of it is comprising the yellow of the sunset which is, however, more red than yellow. I sense I'm going to have more noise in the shadows than I'd like. I think of bjane's clever trick of using a magenta filter which I just happen to have in my pocket.
I slip on the magenta filter. The red channel is unaffected because of course it lets through all the red light, but the blue channel is now worse and the green channel is much worse. I can't increase exposure because the red channel is already as far to the right as it can be without clipping. The filter is not only useless in these circumstances, if I were to use it it would actually decrease dynamic range.
Is this not the case?
[a href=\"index.php?act=findpost&pid=65982\"][{POST_SNAPBACK}][/a]
Here is your daylight exposure, with full exposure to the right. The white square reads RGB 240, 240, 240. Close enough? Reds are to the right. This is about what the RGB histogram of your 5D would show.
[attachment=580:attachment]
Here is the status in the raw channels, which are not shown on your camera's histogram and about which you are unaware. The red is to the left. You can either balance the channels with a filter or in software with a multiplier, as Dennis explains so well with another example. As far as the RGB channel histograms in your camera are concerned, they are not of much help in ETTR without the filter in place, since they show the histogram after the multipliers have been applied. With the filter in place, the multipliers are much closer 1 in the blue and red channels, and the histograms would more nearly represent the raw channels. Another benefit of the filter. What more can I say? In any event, I plan to say no more.
[attachment=581:attachment]
-
Bill,
Would you have any insights about the reason for this response, G>B>R ?
Thanks! Peter
--
[a href=\"index.php?act=findpost&pid=65942\"][{POST_SNAPBACK}][/a]
I believe it is because there are twice as many green pixels in the sensor as red or blue. The sensor is laid out in a square grid so one colour will need to have twice as many pixels as the other two. Green was chosen because the eye is most sensitive to green light.
That's how I understand it anyway. Hopefully someone will correct this if needed.
-
... since both Nikon and Canon have excellent engineers, and there must be an explanation rather than manufacturer stupidity.
Bill
[{POST_SNAPBACK}][/a] (http://index.php?act=findpost&pid=65944\")
Bill,
Maybe it’s more a trade-off rather than an intended purpose. Otherwise above quote could be turned around: if these excellent engineers shaped this response behaviour to G>B>R, then it could be risky to interfere (though I get your reasoning).
Just googled a little bit:
The following reference may not fit perfectly. However, it seems to say that the approach to optimize quantum efficiency can shift the CCD response away from the silicon’s preference for red, towards G > RB. [a href=\"http://www.andor.com/library/digital_cameras/?app=315]http://www.andor.com/library/digital_cameras/?app=315[/url]
Further, this could be promoted or caused alone by the anti-IR-filter in front of the imaging chip. http://www.naturfotograf.com/D200_rev05IR.html (http://www.naturfotograf.com/D200_rev05IR.html)
So yes, a magenta filter seems to make sense (for daylight conditions) provided that there’s enough light to re-balance exposure; which I think was Ray’s point.
Further, I’d say that in a first order such effects of global scaling are independent from the Bayer CFA and the individual R/G/B spectral responses which shape the gamut to form a subset to CIE XYZ (ideally).
http://www.outbackphoto.com/tforum/viewtop...hp?TopicID=1929 (http://www.outbackphoto.com/tforum/viewtopic.php?TopicID=1929)
Willowroot, If the number of green pixel would count, neighbouring green pixel would have to be coupled to produce a higher sensitivity. So the coupled pixel would produce only one signal and wouldn’t be distinguishable anymore. Does that make sense?
Now I should devinitively stop going over my head.
Peter
--
-
You might be gaining something in the highlights (depending on how the white balance exactly works with the channels), but you are also very likely to get a more noisy green channel, since you are doing the very opposite of the recomended expose to the right dogma.
You would expose everything to the right anyway. Suppressing the green channel is only a relative thing, in auto-exposure modes, and calls for a different exposure index, for manual exposure. You certainly wouldn't want to use a filter if you're already struggling with low light; that just weakens channels absolutely, and they all get noisier.
Using a magenta filter like this works well if you want full color over the maximum possible DR. It makes the noise equal in all color channels (assuming Daylight WB), and consequently, the noise is less chromatic.
The normal color response of the cameras almost always has at least one channel that is very weak, and the noise from that channel drags the image down.
-
Exactly! In my example above, comparing a cyan shift corrected in PS as opposed to a cyan shift corrected by use of a magenta filter, one has 2 shots with the same 'external' exposure. The shadows, however, having less 'internal' exposure in the green channel, with use of the magenta filter, will actually be noisier to some degree.
On the other hand, blown detail in the clouds will be brought back slightly. Net effect regarding dynamic range... a big fat zero.
Magenta is just a general correction for daylight WB, assuming white highlights. You can take the idea all the way and filter specifically for the scene; if the blue sky is your highlight, then a red filter will help you push the blue sky closer to RAW max in all three channels, exposing the shadows in all channels as high as possible.
If the cameras had a real RAW RGB histogram, you could do this in the field with a set of filters.
Digital shadows are generally a murky abyss; anything you can do to get all three channels as high as possible is going to increase DR, and decrease readout noise, including horizontal and vertical banding artifacts.
-
Bill,
Would you have any insights about the reason for this response, G>B>R ?
The way I understand it: the sensor, without any filters over it, is most sensitive to red and NIR. The NIR filter cuts NIR and a lot of reds, too, making the red weakest. Blue is already weaker than green because of the sensor's preference for the red end of the spectrum, and the dyes used in the CFA filters have some influence, as well.
These ratios are pretty universal amongst RGB CFA cameras.
Green is considered a reference, blue is generally corrected about 0.95 to 1.5, and red is generally corrected about 1.6 to 2.25 (for daylight).
-
If the cameras had a real RAW RGB histogram, you could do this in the field with a set of filters.
Digital shadows are generally a murky abyss; anything you can do to get all three channels as high as possible is going to increase DR, and decrease readout noise, including horizontal and vertical banding artifacts.
[{POST_SNAPBACK}][/a] (http://index.php?act=findpost&pid=66052\")
If you want something close to a raw histogram with Nikon cameras, just load a white balance into the camera that sets the multipliers to unity--Iliah Borg has such a WB already on his web site:
[a href=\"http://www.pochtar.com/uniwb.nwb]http://www.pochtar.com/uniwb.nwb[/url]
With this WB, the RGB histograms will reflect the raw channels with the gamma correction and tone curve applied. You then set the correct white balance in Nikon Capture. With Adobe Camera Raw, you must use the WB eyedropper rather than a preset WB, since when ACR sees multipliers of unity it thinks it is dealing with a double exposure. You could try setting the red and blue multipliers to slightly less than unity.
-
I slip on the magenta filter. The red channel is unaffected because of course it lets through all the red light, but the blue channel is now worse and the green channel is much worse. I can't increase exposure because the red channel is already as far to the right as it can be without clipping. The filter is not only useless in these circumstances, if I were to use it it would actually decrease dynamic range.
That's why you use a screw-on, or drop-in filter, instead of embedding it inside the camera.
A magenta (or reddish-magenta) is optimal for white highlights (white because of any combination of lighting and subject color). For magenta or reddish highlights, the natural color balance of the camera will be more ideal. However, many camera meters are more sensitive to the red end of the spectrum, so you will most likely not expose to the right in the RAW, anyway, going by the camera's meter. This goes for the magenta filer as well; you need about +1/2 to +2/3 more EC with it (or set the external meter to a lower ISO to compensate).
-
If you want something close to a raw histogram with Nikon cameras, just load a white balance into the camera that sets the multipliers to unity--Iliah Borg has such a WB already on his web site:
Well, I don't have a Nikon so that won't help.
There are still issues, too, I would imagine; the differences between an sRGB and a RAW histogram are not just WB and gamma; the JPEG upon which the histogram is based boosts certain colors, like red flowers. The red in many red flowers gets boosted significantly; the flowers can still clip in the JPEG and histogram, even if there are no WB multipliers, while the RAW red channel is still intact. At least that is what happens with Canon JPEGs.
-
I believe it is because there are twice as many green pixels in the sensor as red or blue.
Actually, that has nothing to do with it. The number of pixels of each color only affects the resolution in that color channel; all three color planes are filled in through interpolation.
Now, if you viewed a diagram of a bayer sensor, or a visualization of RAW data that was the bayer diagram with modulated luminance, then it would appear mostly green for that reason, but the issue here addressed by the filters is one of individual pixel sensitivities, based on color filter.
-
A very easy example to understand is the situation with a tungsten lighted scene. The uncorrected image will have a orange color cast, since the spectrum of the light has much red and little blue. So the blue channel will be exposed purely, the green a bit more and the red a lot more.
Most incandescent lights I've looked at the RAW data from, have almost equal red and green levels for the white square in the GM color checker. Blue is about 1.5 to 2 stops weaker. That's with a Canon, with similar sensitivities as bjanes mentions above.
-
Most incandescent lights I've looked at the RAW data from, have almost equal red and green levels for the white square in the GM color checker. Blue is about 1.5 to 2 stops weaker. That's with a Canon, with similar sensitivities as bjanes mentions above.
[a href=\"index.php?act=findpost&pid=66060\"][{POST_SNAPBACK}][/a]
I looked at a raw conversion with DCRaw on a color checker exposed at 3000K with the Nikon D70 and household bulbs and observed the following values for RGB: 200, 243, 86. If you expose to the right, noise in the blue channel at high ISO is not bad, but with any underexposure, noise increases dramatically. In skin tones, irregular yellow blotches appear where the noise causes the blue channel to drop out--yellow is the complement of blue. Nonetheless, Dennis does make a good point.
An 80A filter would balance the channels, but there is not usually enough light to use the filter and still have decent shutter speed.
-
Well, I don't have a Nikon so that won't help.
There are still issues, too, I would imagine; the differences between an sRGB and a RAW histogram are not just WB and gamma; the JPEG upon which the histogram is based boosts certain colors, like red flowers. The red in many red flowers gets boosted significantly; the flowers can still clip in the JPEG and histogram, even if there are no WB multipliers, while the RAW red channel is still intact. At least that is what happens with Canon JPEGs.
[a href=\"index.php?act=findpost&pid=66058\"][{POST_SNAPBACK}][/a]
I'm sure there is a way to devise a similar white balance for the Canon cameras, but I do not know how to do it.
John is correct about the in camera color processing affecting the histogram with the UniWB. With the Nikon, I think that Mode II is most neutral. One should use normal saturation. Stay away from Mode III (the Velvia mode)
-
Ray, how do you meter a scene, if you want to measure only the light, and want to bann all reflective issues and influences of colors to the meter? If you would use a grey card, then the magenta filter method would be fine with you.
Dennis,
I take pictures principally because I enjoy the extroverted activity of getting around and capturing the moment. I don't even own an external light meter. I have no commissions to produce catalogues of Dulux paints. Nor am I a fashion photographer who might worry about getting the precise shade and hue of a particular dress just right. Color accuracy for me is only important as far as it's believable and/or pleasing. A bit of exaggeration is fine.
I rely almost exclusively upon my camera's light meter and feed back from the camera's RGB histogram. Since the histogram is based upon a jpeg conversion, I've changed its behaviour by setting the contrast to a minimum in landscape (picture style) mode. This has no bearing on the RAW data of course, but it helps me to get a clearer idea of a true exposure to the right when the scene has a high DR. A small amount of highlight warning (blinking) in the brightest part of the sky, for example, is just right. This contrast setting in landscape mode, however, is not appropriate for other types of photography, such as use of flash indoors. The histogram then seems to produce a false impression that more exposure is required. I need to do more experimentation with different contrast settings to get a better idea of which setting produces the most accurate histogram feed-back under what conditions.
Colors of the subject play no role at all. It's all about adopting different sensitivities of the channels of a certain sensor to a certain light (temperature). A very easy example to understand is the situation with a tungsten lighted scene. The uncorrected image will have a orange color cast, since the spectrum of the light has much red and little blue. So the blue channel will be exposed purely, the green a bit more and the red a lot more. If you want to handle this problem with software based white balance only, you have to increase the values of the blue channel exessively and the green ones a bit less to match the red channel values in order to get a neutral grey, right? So you will introduce much noise in the final image. Therefore, it would be a way better idea, to balance the light before it reaches the sensor, eg using a blue 80B filter.
Well, I wouldn't go as far as to say the colors of the subject play no role at all, but I see what you are getting at and it seems to be in line with what I've been saying. You are recommending use of a filter to suit the lighting conditions of the external scene, which sounds perfectly sensible to me.
What I'm dubious about is the practice of choosing a filter to suit the internal characteristics of the camera's RGB channels, without reference to the lighting conditions of the external scene. In your above example you've described a DR benefit resulting from use of a blue filter to suppress the warm (yellowish) glow of tungsten lighting. Couple that with a magenta filter to correct the imbalance of the camera's RGB channels and you have, in effect, an even stronger blue filter which I guess you might call purple, if there's still some red coming through.
The above proposed method is all the same, it balances light to the sensors needs. This balancing has nothing to do with colors in the scene, like green foliage or red skyes. It only has to do with the light itself and the sensor. So you'd need a different setup for each sensor and light temperature. If that's worth the hassle, is an individual question.
I'm having trouble with the logic of the above statement. Unless you are pointing the camera directly at a light source, the camera is recording light that has been reflected off your subject and which has therefore been 'modified' in some way by the 'light reflecting' characteristics of the subject. A 'light reflecting' characteristic of a surface is usually described as its 'color'. Instead of fitting a blue filter to reduce the amount of red and green entering the camera, you could paint the subject an appropriate color which could have the same effect (if it was done properly). Of course that's totally impractical in most situations. I just mention it to illustrate the point that the light entering the camera is an interaction between the nature of the light source and the nature of the surface reflecting the light. It therefore doesn't seem to be true to say This balancing has nothing to do with colors in the scene, like green foliage or red skyes.
The other question is, if you want the light to be balanced. A beautifill colored sunset will loose everything, if you match the WB exactly to the temperature of light. So sometimes a color cast is exactly what you want.
Now we're getting to the crux of the matter. Of course you don't necessarily want to correct every scene on the basis that any light source that is not neutral is bad, and a sunset is a perfect example.
Consider the following image, taken with a 5D without use of filters of any sort. I'm struggling to understand how a magenta filter would have helped this scene. As far as I see, such a filter would merely make the blues and greens noisier, pushing them more to the left of the histogram. As it is, the scene is way beyond the dynamic range capabilities of the camera so any filter that could have increased the DR of the camera would have been appreciated.
I've also included the imaage in the ACR window showing the settings and histogram plus an enlargement of the histogram to help some of you experts advise me on appropriate use of filters in this situation.
[attachment=582:attachment] [attachment=583:attachment] [attachment=584:attachment]
But as a matter of principle, the filter method is absolutely correct. It's always a better idea, to balance the light before it hits the sensor, rather than balancing it afterwords by software.
I agree with the principle, but I'm dubious about the practice.
ps. That's odd. The shadow detail of the above processed image is better in PS than it is here. I'm including a lightened version.
[attachment=588:attachment]
-
The other question is, if you want the light to be balanced. A beautifill colored sunset will loose everything, if you match the WB exactly to the temperature of light. So sometimes a color cast is exactly what you want.
But as a matter of principle, the filter method is absolutely correct. It's always a better idea, to balance the light before it hits the sensor, rather than balancing it afterwords by software.
The actual recording of RAW data doesn't have to be filtered for the WB or desired WB, though, for optimum quality. You can always darken channels in the conversion, to bring them where you want them. The best recording is had when the filtration treats each color channel like three B&W exposures that you are trying to each expose to the RAW right.
Of course, this is idealistic, as many RAW converters are incapable of rendering filtered images properly. ACR just shrugs its shoulders when I click the WB tool on some "white" things with strong filters, and ignores my attempt. Each program has its own limits. I really don't see why RAW converters don't sport totally linear exposure compensation, or free white-balance, in a sort of "pre-gain" stage, because there is nothing simpler than white-balancing and scaling (exposure compensating) linear RAW data. Subtract the blackpoint, and multiply/divide.
A CC30M works fine in ACR, though, in daylight. It is not a perfect correction, though. I have put filters over my 550EX flash that were almost a perfect RAW WB for the white GM CC square, but it seemed to be just barely outside what ACR would balance, at least a couple versions back, anyway.
-
I looked at a raw conversion with DCRaw
Are you the guy I was talking to about the problems with DCRAW in the old Galbraith forum? I've noticed that DCRAW has sprouted a new unaltered RAW command line option, "-D". Did you have something to do with that?
Still has a bit of a problem, though. PS can't import the output of DCRAW without posterizing it. When DCRAW -s set to output PSD files, it really should scale the data so that PS reads it in as exactly 0 to 4095 (that may simply mean multiply by two, or might have some gaps). If you are friendly with D. Coffin, you might suggest that to him.
-
I rely almost exclusively upon my camera's light meter and feed back from the camera's RGB histogram. Since the histogram is based upon a jpeg conversion, I've changed its behaviour by setting the contrast to a minimum in landscape (picture style) mode.
[{POST_SNAPBACK}][/a] (http://index.php?act=findpost&pid=66075\")
Nice shot, Ray. Since the white balance affects the color histograms, one must take the white balance into consideration when interpreting them. I assume the WB was daylight. The red channel appears blown, but since the red multiplier is nearly 1.92, the channel may not actually be blown. The actual raw red histogram would be shifted one stop to the left. However, I doubt that you could push exposure significantly without blowing the red channel. It would be helpful if you could post the raw file or the raw histogram from a DCRaw conversion.
Consider the following image, taken with a 5D without use of filters of any sort. I'm struggling to understand how a magenta filter would have helped this scene. As far as I see, such a filter would merely make the blues and greens noisier, pushing them more to the left of the histogram. As it is, the scene is way beyond the dynamic range capabilities of the camera so any filter that could have increased the DR of the camera would have been appreciated.
I've also included the imaage in the ACR window showing the settings and histogram plus an enlargement of the histogram to help some of you experts advise me on appropriate use of filters in this situation.
[a href=\"index.php?act=findpost&pid=66075\"][{POST_SNAPBACK}][/a]
I don't consider myself an expert in the use of filters in this situation, since I've never used them in this context and my reasoning is purely theoretical. However, I will venture a few comments to stimulate discussion. Since the magenta filter gives only about 2/3 stop more dynamic range, it obviously would not solve your problem.
Examination of the individual channels shows that shadow noise is actually worst in the red, and unfortunately you can't increase exposure much in that channel. Surprisingly, the blue channel appears relatively clean. The green channel has noise in between the other two channels.
It would seem to me that a cyan filter might be of some use by holding back some of the red and allowing you to increase overall exposure and balance the channels so as to improve DR and noise in the blue and green. Unfortunately, without raw histograms it would be difficult to judge what strength of filter to use or whether some green needs to be held back as well in order to balance the channels.
Since filters would give less than 1 stop additional dynamic range and the scene may be fleeting, I do not think they would be a practical solution.
Another way to increase dynamic range is to use ISO 100. The effect of ISO on DR is shown in a table on Roger Clark's web site.
[a href=\"http://www.clarkvision.com/imagedetail/evaluation-1d2/index.html]http://www.clarkvision.com/imagedetail/eva...-1d2/index.html[/url]
-
For magenta or reddish highlights, the natural color balance of the camera will be more ideal. However, many camera meters are more sensitive to the red end of the spectrum, so you will most likely not expose to the right in the RAW, anyway, going by the camera's meter. [a href=\"index.php?act=findpost&pid=66057\"][{POST_SNAPBACK}][/a]
Hi! John,
Are you a refugee from the RG forum, or do you still post there? I find I have enough to keep me occupied here, so haven't subscribed.
The two DSLRs I've used most, up to date, are the Canon D60 and 5D. The 20D was a mistake. If I'd known the 5D was on its way, I wouldn't have bought it.
Both the D60 and the 5D appear to have a weak blue channel which clips before green. This results in a cyan shift in the brighter parts of a blue sky when overdoing it with ETTR. The 5D is not as bad as the D60 in this respect, so I would deduce that the blue channel in the 5D is slightly stronger (higher DR in relation to the green channel) than in the D60. Right?
The red channel in the D60 was at least as bad as the blue channel, showing a similar effect with the yellow glow of sunsets. The red component of yellow would clip first, leaving a predominance of green. Green in a sunset? Not on.
This is not the case with the 5D. The red channel appears to be at least as strong as the green channel. If I were to use a filter on the 5D to equalise the sensitivity of the channels, it would surely be a blue filter to bring both the red and green channels down to the level of the weaker blue channel.
-
Another implication of this use of filters to equalise the camera's channels, which I can't quite fathom, is the 'lowest common denominator' effect.
Bjanes has come up with a multiplier of 1.92 to get the sensitivity of the red channel equal to that of the green. (Probably applies to the D60 or 10D, but not the 5D). Sensitivity has a lot to do with dynamic range. If you effectively 'disable' the brighter channels to correspond with the 'noisier' channels, you have effectively reduced dynamic range in relation to a completely neutral or B&W scene, or an imaginary scene where every color and hue consists of equal parts of R,G&B.
In other words, with a D60 for example, if I want to convert a color image to B&W, I will get the best dynamic range by using just the green channel. Is this not the case?
-
Another implication of this use of filters to equalise the camera's channels, which I can't quite fathom, is the 'lowest common denominator' effect.
Bjanes has come up with a multiplier of 1.92 to get the sensitivity of the red channel equal to that of the green. (Probably applies to the D60 or 10D, but not the 5D). Sensitivity has a lot to do with dynamic range. If you effectively 'disable' the brighter channels to correspond with the 'noisier' channels, you have effectively reduced dynamic range in relation to a completely neutral or B&W scene, or an imaginary scene where every color and hue consists of equal parts of R,G&B.
In other words, with a D60 for example, if I want to convert a color image to B&W, I will get the best dynamic range by using just the green channel. Is this not the case?
[{POST_SNAPBACK}][/a] (http://index.php?act=findpost&pid=66090\")
I know this thread is quite long and convoluted, but if you had read my posts, the multiplier of 1.92 is for the Canon EOS 1D Mark II. The multipliers do vary somewhat between camera models, but tend to be in the same ball park range. For example, see Julia Borg's data for Nikons:
[a href=\"http://www.pochtar.com/NikonWhiteBalanceCoeffs.htm]http://www.pochtar.com/NikonWhiteBalanceCoeffs.htm[/url]
To white balance, one makes the output of the channels equal for white and grays (e.g. the neutral squares of the color checker). One does not equalize non-neutral channels except for special purposes, in which case the image is no longer white balanced. If you want to determine the blue multiplier for your 5D, you would have to examine the linear raw values of the channels when a neutral target is exposed under daylight conditions.
Dynamic range is defined as linear full well (electrons)/read noise (electrons). In a daylight scene with full ETTR and daylight white balance, only the green channel gets to full well. The red and blue channels will be less than full well unless you have used balancing filters and will have less effective DR. You would get the best DR by using the green channel, but the tones would be rather strange. Just look at the green channel of your images in Photoshop. It would best to equalize the channels with filters.
-
Are you the guy I was talking to about the problems with DCRAW in the old Galbraith forum? I've noticed that DCRAW has sprouted a new unaltered RAW command line option, "-D". Did you have something to do with that?
Still has a bit of a problem, though. PS can't import the output of DCRAW without posterizing it. When DCRAW -s set to output PSD files, it really should scale the data so that PS reads it in as exactly 0 to 4095 (that may simply mean multiply by two, or might have some gaps). If you are friendly with D. Coffin, you might suggest that to him.
[a href=\"index.php?act=findpost&pid=66080\"][{POST_SNAPBACK}][/a]
John,
Yes, I do recall having a discussion with you on this topic. I think it was concerning white balance, and you did bring up some good points.
The -s switch if for Fuji cameras and is ignored for other cameras. Did you mean the -3 switch? I did try the -D switch, but the data are not scaled for photoshop display and appear quite dark. I usually use the -m -n -3 switches in the older version, which multiply the raw values by 16, converting 12 bit notation to 16 bit. I think that there are still some scaling problems and some clipping occurs. Julia Borg advises the -b 0.25 switch, but the output is too dark for my taste and you would have to multiply the pixel value by 4. I have not really used the latest version that much.
-
Most incandescent lights I've looked at the RAW data from, have almost equal red and green levels for the white square in the GM color checker. Blue is about 1.5 to 2 stops weaker.
...following values for RGB: 200, 243, 86.
Thanks for correction and clarification. But I was dealing more with the general issue, than with exact weighting of each cahnnel. But as orange is a mixture of red and green, it makes perfect sense, that the green channel is more up to the red, rather than to the blue.
In your above example you've described a DR benefit resulting from use of a blue filter to suppress the warm (yellowish) glow of tungsten lighting. Couple that with a magenta filter to correct the imbalance of the camera's RGB channels and you have, in effect, an even stronger blue filter which I guess you might call purple, if there's still some red coming through.
STOP! I think, here is a misunderstanding! I did not mean to stack the magenta filter to the blue filter. There is ONE light temperatur (maybe with cast in the purple-green-axis? But let's ignore that for simplification), where a specific sensor captures exactly equal values in all three channels in the Raw file. For example in the lower picture of this posting from bjanes (http://luminous-landscape.com/forum/index.php?showtopic=10937&view=findpost&p=66005), you'd have a value of (180,180,180)*. Now, according to the color of your light source, you have to filter it with a special CC filter. A blue one for Tungsten light and a magenta one for day light, simplified. I did not mean to use the magenta filter in any case as kinda standard filter.
I'm having trouble with the logic of the above statement. Unless you are pointing the camera directly at a light source, the camera is recording light that has been reflected off your subject and which has therefore been 'modified' in some way by the 'light reflecting' characteristics of the subject
Correctly, and that's exactly what you want to eliminate, if you measure the light using a grey card.
Instead of fitting a blue filter to reduce the amount of red and green entering the camera, you could paint the subject an appropriate color which could have the same effect (if it was done properly).
No, that wouldn't change the light, only the object color. What you want to do, is to calibrate the sensor to the light color, so that it renders any object perfectly w/o regarding its color. That's what a part of CM is about: To neutralize the influence of different light colors to get to the 'real' colors. Again, if that is necessary or not, depends on personal or professional needs.
It therefore doesn't seem to be true to say This balancing has nothing to do with colors in the scene, like green foliage or red skyes.
An imagier works perfectly, if a grey card would give equal amounts of load in each colored pixel, right? So, a grey is a grey, and a red is a red. But that's not the case, as you can see in bjanes pictures: For the sensor, a grey is a greenish color with unequal amounts of 'light' in each channel. And so is a red not a red, but biased in the same way. So, you now want to balance the light, in order to get equal amount of load in each photo site. That's, where a CC filter comes in: If it's the right color, it just blocks the right amount of light of the accordingly color, so that each photo site get's the same amount of light. If this is done, a grey is a grey and a red is a red again. It does not depend on obejct color, but only on light color. This is the balancing, I am talking about.
Consider the following image, taken with a 5D without use of filters of any sort. I'm struggling to understand how a magenta filter would have helped this scene.
A nice picture! But strange blue? Anyway, you didn't get the point: A magenta filter would be useful in daylight condition at about 5500 K or so. This scene here has a much lower light temperature, more like 3500 K or so (just guessing). So you'd need another filter (more like the 80B), if you ever wanted to correct the light here, what's clearly not the matter, since the scene 'lives' from the wonderful warm light.
As it is, the scene is way beyond the dynamic range capabilities of the camera so any filter that could have increased the DR of the camera would have been appreciated.
You can see the problem with this image in your ACR conversion: Massive and ugly noise banding. This is due to the fact, that the blue channel is poorly exposed, and so has to be amplified strongly, which brings up the noise. To get the perfect result, you had to do the following: Measure the light temperature** and placing the accordingly filter in front of the lens. Of course, the exposure time might then come to an unacceptable range - bad luck! But if it worked, you had a 'perfectly' balanced pictured without the beautiful warm cast, more like a boring day light picture. But that picture would suffer much less noise problems. What you had to do now, is to change the color temperature in ACR, the get the beautiful warm color cast again. And that's, were the ultimate trick is: Now, you have to reduce the blue channel. This is not critical regarding noise. Before, you had to bring up the blue channel, and this is very critical with noise!
So I agree perfectly with you, that it's very impractical to meter the light temperature each time and choose the right filter out of hundreds. But I think something like "use a 80C filter for sunsets" could bring the sensor data in the right place, and would be very practicable. It's up to you to verify, if this could be useful for your work.
Best Regards,
Dennis.
* Wouldn't it be interesting, to figure out this light temperature? Has it anybody done so far?
** It actually is very simple to measure the light temperature with a digital camera: You do it every time you use a manual WB. But unfortunately, the camera only uses the computed settings for internal Raw conversion, rather than showing the values on the display. Or is there meanwhile a camera, which does it?
-
The other question is, if you want the light to be balanced. A beautifill colored sunset will loose everything, if you match the WB exactly to the temperature of light. So sometimes a color cast is exactly what you want.
[{POST_SNAPBACK}][/a] (http://index.php?act=findpost&pid=66002\")
I usually use daylight color balance for sunsets to preserve the red glow, but Bruce Fraser had another suggestion on the Adobe Camera Raw forum. In that thread sunset colors were also being discussed:
[a href=\"http://www.adobeforums.com/cgi-bin/webx?7@@.3bb6a85c.3bbfb49a/3]http://www.adobeforums.com/cgi-bin/[email protected]/3[/url]
He suggests illuminating a MacBeth color checker with indirect sunlight and use R4C2 for white balance. If you don't have the CC with you, he suggested taking a reading from whitecaps (assuming you are on the sea shore). The method is counterintuitive to me, but worth a try. Bruce is not into magenta filters.
-
It would seem to me that a cyan filter might be of some use by holding back some of the red and allowing you to increase overall exposure and balance the channels so as to improve DR and noise in the blue and green. Unfortunately, without raw histograms it would be difficult to judge what strength of filter to use or whether some green needs to be held back as well in order to balance the channels.
Well, I would agree with that analysis, bjanes. This illustrates the point I've been trying to make all along. A cyan filter has roughly the opposite effect of a magenta filter. As I see it, if it were possible to select the appropriate strength of cyan filter, (and without a RAW histogram and a bag full of filters to choose from, that would be extremely difficult, but with a lot of experimentation one might gain sufficient experience to get a 'feel' for the right filter), then a cyan filter would definitely allow a reduction in noise in the blue and green channels, with increased exposure, but not in the red channel if that channel is already exposed to the right.
However, there are no pure reds in the image, as far as I know. Everything is a mixture of red, green and blue, so if the noise is reduced in just 2 of the 3 channels, a worthwhile reduction of noise should be noticeable in the whole of the shadow area, which is substantial in this image.
However, I accept that this is an extreme example and there would be many situations where use of a magenta filter would help dynamic range, but it would be a varying result; sometimes perhaps a 2/3 stop increase; sometimes perhaps a 1/3rd stop increase and sometimes no increase at all.
It seems to me that to get the maximum dynamic range that is possible from a sensor, one could theoretically choose the appropriate filter or combination of filters so that there will always be 'some' wells in all three channels completly full (and overflowing in the case of specral highlights). The custom WB should then contain the ratios by which certain channels need be reduced to get the required balance and temperature.
If I've understood this process, it should be possible to stick any type of color filter in front of the lens, whether it's appropriate or not, and make a custom WB that will bring things back in line during conversion. If this is true, then the filter that is the most appropriate should be the one that allows the greatest increase in exposure without suppressing the weakest channel (in terms of the light entering the lens), and without clipping any of the channels.
-
STOP! I think, here is a misunderstanding! I did not mean to stack the magenta filter to the blue filter. There is ONE light temperatur (maybe with cast in the purple-green-axis? But let's ignore that for simplification), where a specific sensor captures exactly equal values in all three channels in the Raw file.
Dennis,
Whatever the WB and whatever the tempaerature of the light source, the sensor cannot capture equal (maximum) values in all 3 channels unless equal (maximum)values of each primary color are passing through the lens. If I'm photographing a room, for example, where eveything in it is a shade of purple (another extreme example to illustrate my point) there's no chance the sensor will captures equal values in the green channel. In this situation you would need to use a green filter to suppress the red and blue light passing through the lens, take a shot (with green filter on) of a white (or grey) card which is illuminated by the light source rather than the reflection from the purple surfaces in the room, in order to get your custom WB. You could then increase exposure till the small amount of green in the scene reaches a maximum value without clipping. By doing this, you would preserve the levels of red and blue which you would have got without green filter, but reduce noise of the small amount of green that is in the scene quite substantially. Is this not the case?
-
Whatever the WB and whatever the tempaerature of the light source, the sensor cannot capture equal (maximum) values in all 3 channels unless equal (maximum)values of each primary color are passing through the lens.
It doesn't matter, what's entering the front lens, but what's hitting the sensor. You take care, that the light entering the front lens should represent a neutral grey tone, eg. by focusing to a grey card. The light, which reaches the sensor, should have equal amounts of all three primaries, ideally. If the light entering the front lens has another composition (color cast), you correct it using a CC filter. So the reflected light from a grey card, which has some unbalanced spectrum due to 'not ideal' light source, will be balanced by the CC filter. There is a difference between an unbalanced light caused by the color of the object or caused by color of the light itself. The latter, you try to eliminate with the CC Filters. The first is, what you actually want to capture.
If I'm photographing a room, for example, where eveything in it is a shade of purple (another extreme example to illustrate my point) there's no chance the sensor will captures equal values in the green channel. In this situation you would need to use a green filter to suppress the red and blue light passing through the lens, take a shot (with green filter on) of a white (or grey) card which is illuminated by the light source rather than the reflection from the purple surfaces in the room, in order to get your custom WB.
In this case, you should avoid reading a grey card, which would be illuminated by the reflected light of your purple room. You should measure the light source directly, instead.
Dennis.
-
It doesn't matter, what's entering the front lens, but what's hitting the sensor.
Dennis,
First let me say I am not speaking from a wealth of experience creating custom white balances. I'm really just trying to understand the principle and what you are saying simply doesn't make sense to me.
Whatever hits the sensor must have passed through the lens, except internally generated noise from heat sources etc. Whether or not the WB is done properly; whether or not you've picked up too much reflected light from strongly colored surfaces in the scene, the WB process (as I understand it, and correct me if I'm wrong) is an information tag which basically says, 'this heavily green biased part of the image, which the spot meter has focussed on, is supposed to be white. The green channel must be reduced accordingly throughout the entire image during conversion.'
The white balance is thus an automatic 'levels adjustment' set of instructions. If you get it wrong, you then have to manually adjust levels in PS, or use the grey balancing eyedropper, whatever.
You take care, that the light entering the front lens should represent a neutral grey tone, eg. by focusing to a grey card.
I don't think so. You take care that the light before it enters the lens should represent the temperature of the light source. Adding any color filter to the lens, for whatever purpose, effectively changes the temperature of the light source. But that is the whole purpose of adding a filter, isn't it. You want to change the temperature of the light source in such a way that the light, after passing through the filter and lens, will either (1) contain at least some saturated levels in each channel for the purpose of increasing DR, or (2) already contain a neutral balance in respect of a white or grey surface (in the case of your suggestion of a blue filter with a tungsten light source).
I don't see why we should be making a distinction between adding a magenta filter to correct for the internal imbalance of sensor sensitivity as opposed to adding a different filter which corrects for both imbalances, internal and external, unless of course we get ourselves into a situation where we are outside of the range of the camera's custom WB adjustments, in which case we are stuffed.
I think I have a red filter somewhere. Perhaps it's time for me to experiment with creating some custom white balances under differernt conditions using a red filter. I might afterwards have to confess I'm totally wrong and confused .
-
You take care that the light before it enters the lens should represent the temperature of the light source. Adding any color filter to the lens, for whatever purpose, effectively changes the temperature of the light source.
Perhaps I should add clarification here. From the camera's perspective, that is from the WB algorithm's perspective, it makes not one whit of difference whether the white card is tinged with blue light as a result of the light source having a high temperature, or whether the light, as reflected from the white card, has a high temperature because it has passed through a blue filter on the lens, on it's way to the sensor. The WB system is not aware of the reason for this excessively high temperature. All it knows is, that which is supposed to be white (the white card) is in fact slightly blue, and it makes levels adjustments accordingly by reducing the blue channel in the entire image. Is this not right?
Well, I certainly hope all this confusion has not given you a high temperature. I've felt like taking a few asprins whilst trying to sort through this issue .
-
Hello,
That is a very interesting discussion.
I've come up with an idea that might be not very practical but is it valid from theoretical point of view?
Assuming the shots in daylight
1. We take a shot with ETTR exposure - this will give green chanel at full well in linear RAW
2. Take a second shot with +1EV - this will push green to overexpose, but get blue at full well in linear RAW
3. Take a third shot with +2EV - overexposing green and blue but getting red at full well in RAW.
4. Process the three shots placing the corresponding chanels a the right with white balance tool (or exposure).
5. Combine GREEN, BLUE and RED chanels from the three shots into one image.
Will this work?
Allan
-
Whatever hits the sensor must have passed through the lens,
Yes, you are right. I meant it in a different way. Obviously the combination of tints of the CFA and the wave length dependant sensitivity of the imagier leads to an unbalanced channel situation, even if you sent 'balanced' light (whatever that means) through the lens. The balance is up to your light source. A grey card lit with D50 light will give a different reading, than if lit with D65 light. There is supposed to be one temperature, say eg. 5,800 K, where the reflected light of a grey card produces exactly evenly filled color channels. If you are variing the light temperature, the distribution will change in one or another direction.
So the concept here is, you first have to analyse using a grey card, what the current lighting does to the three primary channels. Then you can compensate a color cast using a CC filter, and thus manage to get evenly filled channels. This will allow to do ETTR to all channels perfectly, and not - as shown - to bring only one channel to the right, and waste some headroom with the other two channles. Okay, if the hassle is worth the effect, is another question. With daylight, an advantage of 2/3 f stop was reported. Well, that's not that much. But maybe this technique may help to avoid blowing the green channel - don't know. But with tungsten light, this filtering definately makes sense, since here the difference between the channels is way bigger.
The white balance is thus an automatic 'levels adjustment' set of instructions. If you get it wrong, you then have to manually adjust levels in PS, or use the grey balancing eyedropper, whatever.
Yes. But it's preferrable, not to have to do this by a large amount, so it's better, if the light is already balanced before it hits the sensor (eg via using a CC filter). And please be aware, I am talking NOT about object colors, but the light color (grey card reading).
You take care that the light before it enters the lens should represent the temperature of the light source.
Yes, that's what I said. That's, what you'd use a grey card for or some white target.
I don't see why we should be making a distinction between adding a magenta filter to correct for the internal imbalance of sensor sensitivity
This is your misunderstanding: There is no 'internal imbalance'. There is only an imbalance between the light and the sensor. The above mentioned magenta filter wokrs only with this specific light temperature. It's exactly the same as with film. Think about. There is no such thing as 'external imbalance'. There are only two factors:
- the spectral sensitivity of the media (film or digital device)
- the spectral distribution of the light
To match these two, you need one filter. Taht's all.
All it knows is, that which is supposed to be white (the white card) is in fact slightly blue, and it makes levels adjustments accordingly by reducing the blue channel in the entire image. Is this not right?
That's perfectly right. But it's better to balance the light, than to tweek the channels via WB.
Take a picture under 'heavy' tungsten light, like a candle or very dimm light bulb, the camera set to JPEG and daylight. Then add a blue filter and take another shot. Try to WB both images and compare. You'll end up with massive problems with the first image. Of course, using Raw the problem is way less significant, but it's not gone.
Dennis.
-
Will this work?
It should. Go and evaluate it. But be aware, that it's not always possible to make three shots. The same problem like with HDR scenes: Can you bracket and use DRI technique or should you use a gradient filter?
Dennis.
-
Hello,
That is a very interesting discussion.
I've come up with an idea that might be not very practical but is it valid from theoretical point of view?
Assuming the shots in daylight
1. We take a shot with ETTR exposure - this will give green chanel at full well in linear RAW
2. Take a second shot with +1EV - this will push green to overexpose, but get blue at full well in linear RAW
3. Take a third shot with +2EV - overexposing green and blue but getting red at full well in RAW.
4. Process the three shots placing the corresponding chanels a the right with white balance tool (or exposure).
5. Combine GREEN, BLUE and RED chanels from the three shots into one image.
Will this work?
Allan
[{POST_SNAPBACK}][/a] (http://index.php?act=findpost&pid=66248\")
Allan,
I do not see why your method would not work if anyone wanted to take the trouble to carry it out. However, your exposure factors are off a bit. Let us use a Nikon D200 for the exercise, since channel multipliers for white balance are readily available for this camera.
[a href=\"http://www.pochtar.com/NikonWhiteBalanceCoeffs.htm]http://www.pochtar.com/NikonWhiteBalanceCoeffs.htm[/url]
In the experiment, the green channel is already balanced. For the red, we have to multiply the exposure by 1.83. In terms if f-stops this is Log base 2 of 1.83 or 0.87 stops. For the blue the multiplier is 1.36 or 0.44 stops.
If you are using Photoshop with ACR and are bracketing exposures, it might make more sense to bracket more widely and use HDR. You would get far more dynamic range with HDR than the use of filters.
Bill Janes
-
This is your misunderstanding: There is no 'internal imbalance'. There is only an imbalance between the light and the sensor. The above mentioned magenta filter wokrs only with this specific light temperature.
Dennis,
Are you sure about this? On the previous page, there's an explanation from John Sheehy, who seems to be heavily into such technical matters, which implies there's a sensitivity imbalance amongst the 3 channels. Refer to the following quote.
The way I understand it: the sensor, without any filters over it, is most sensitive to red and NIR. The NIR filter cuts NIR and a lot of reds, too, making the red weakest. Blue is already weaker than green because of the sensor's preference for the red end of the spectrum, and the dyes used in the CFA filters have some influence, as well.
These ratios are pretty universal amongst RGB CFA cameras. Green is considered a reference, blue is generally corrected about 0.95 to 1.5, and red is generally corrected about 1.6 to 2.25 (for daylight).
Since D65 is such a commonly used standard, one wonders why Canon does not include a magetna-ish CC filter, perhaps immediately under the sensor's protective covering, especially since their recent models have such excellent performance at high ISOs, or better still, why do they not simply increase the density of the blue and green filters in the filter array so that the sensor is calibrated more closely to D65?
I think I understand the implication of the magenta filter. Because the D65 WB 'tag' will always try to increase the red channel by a factor of, say 1.92 in respect of the green channel, we can never expose the red channel fully to the right because the application of the WB levels adjustment (during conversion of RAW image) will then blow the channel. If we create the conditions for a custom white balance that doesn't apply this automatic increase of levels in the red channel, by using a magenta-ish color correcting filter when creating the custom WB, then we are home and dry. We are then in a situation of being able to increase exposure by up to a full stop or so, apparently in all situations, resulting in a less noisy red channel and perhaps a slightly less noisy blue channel, depending on the camera model.
But I still think that any increase in DR will be dependent on the composition of the shadows. Shadows which consist of lots of magenta will benefit most and shadows which are predominantly green and cyanish will benefit the least in respect of any increase in DR, the reason being, if you suppress the green and (maybe) blue channels, the additional exposure simply brings the green and (maybe) blue channel back to the same level they would have been without use of a filter. One is relying mostly on a reduction of noise in the red channel to get any real increase in dynamic range. Since there's always going to be at least some red in the shadows, I shall have to concede that use of a red/magenta filter is going to be of some benefit to DR in most high dynamic range scenes.
However, I doubt that the benefits would be worth the hassle for me. If there is a way of extending the principle by using the camera's WB adjustments to correct levels after using stronger filters that match the general color composition of the scene, I think that could result in substantial increases in DR.
-
5. Combine GREEN, BLUE and RED chanels from the three shots into one image.
Will this work?
Allan
It will work, but you have to do it RAW to really get it right, as demosaicing and RGB color conversion cross-contaminate colors. Then, you have to get this data back into a converter. I have been thinking about this, and I know how to do it, but I haven't getten around to it yet, because I have to do some programming and I don't program all the time, so I get a bit rusty, and time is limited.
There is also the issue of registration; you need for the camera not to move between shots, otherwise you will be getting a blurred, under-sampled image. Would work best with super-wide angle lenses on sturdy tripods.
-
However, I doubt that the benefits would be worth the hassle for me. If there is a way of extending the principle by using the camera's WB adjustments to correct levels after using stronger filters that match the general color composition of the scene, I think that could result in substantial increases in DR.
That would be better yet. The idea of the reddish magenta filter is that it brings the sensitivity of the camera, as a system, to a natural daylight WB. That is ideal if you want full color at the highest nearly-neutral highlights (like pastels in the sun, or subtle color differences on white clouds), and as deep as possible into the shadows. If your highlight is something else, and you want full color and DR, then the optimal filtration might be something else.
The bottom line is that you will get the best RAW capture when you can expose each of the 3 RAW color channels to the right, independently.
-
Dennis,
Are you sure about this? On the previous page, there's an explanation from John Sheehy, who seems to be heavily into such technical matters, which implies there's a sensitivity imbalance amongst the 3 channels. Refer to the following quote.
Okay, Ray, you are right here. Given a perfect balanced light source with an absolute evenly distributed spectrum, you can record such an absolute 'internal' imbalance. You always can divide the total imbalance into two parts: One 'internal' imbalance between the sensor and a standard (like the above perfect light or D65), and the 'external' imbalance between the given light and the chosen standard. So you had to filter the whole thing in two steps: First calibrate the sensor to the standard, second match the light to the standard. This might be not very effective. It's much easier to see the imbalance as one 'piece'. So the only thing, which matters is, how records the sensor the actual light?
-
That would be better yet. The idea of the reddish magenta filter is that it brings the sensitivity of the camera, as a system, to a natural daylight WB. That is ideal if you want full color at the highest nearly-neutral highlights (like pastels in the sun, or subtle color differences on white clouds), and as deep as possible into the shadows. If your highlight is something else, and you want full color and DR, then the optimal filtration might be something else.
The bottom line is that you will get the best RAW capture when you can expose each of the 3 RAW color channels to the right, independently.
[a href=\"index.php?act=findpost&pid=66313\"][{POST_SNAPBACK}][/a]
What John says makes sense, but it is hard enough to expose to the right when using normal color balance and no filters. The histograms and blinking highlights often underestimate how far to the right the exposure is.
If you are shooting in daylight, using filters and a custom white balance and photographing a full spectrum subject with neutral (gray-white) tones that must be neutral in the image, then daylight white balance is best. The magenta filter is only an approximation. Consider the this image of the color checker. The raw histogram of the straight shot without filters is shown:
[attachment=596:attachment]
If you add magenta filtration to bring the green down to the level of the red channel, the blue is a bit too strong and some yellow filtration is needed to bring the blue into balance. I calculate that CC26M and CC13Y would be needed for full correction with +0.87 EV additional exposure (for the Nikon D200). Most would just settle for approximate balancing with a magenta filter.
For other situations, such as Ray's sunset, one should expose all the channels to the right as John suggests, but selection of the proper filters would be no simple matter. The camera color histograms would not be helpful unless the camera white balance is set such that the channel multipliers are all 1.0 (UniWB for Nikons). This is a lot of bother for a very modest improvement in dynamic range and I predict that very few photographers would take the trouble. Personally, I would use the lowest possible ISO and skip the filters for the sunset.
The magenta filter is a reasonable compromise for full range daylight illuminated subjects.