I shoot dance and theater and the dynamic range of the scenes can be a challange for the meter of my 1DmkII. I've experimented with all four metering modes and I've found that in the changing light conditions of the stage that more often than not, evaluative works the best. I do have to dial in a bit of exposure compensation to keep the highlights from blowing out, but it can vary based on the costume of the subject and the lights. Lots of hard light with light skin and/or light costumes need more negative exposure comp than a less contrasty scene. Makes sense, but hard to do in practice. At the high ISO's I shoot, I don't want to underexpose which will accentuate the noise, but I definitely don't want to overexpose.

But over the weekend as I was shooting for a local non-profit dance studio, it dawned on me that a very useful variation of the evaluative metering mode would be a 'no-clipping' option. The camera would evaluate all the metering zones and then based on the measurements calculate the exposure such that the brightest zone would be just short of clipping, or 'exposed to the right'.

It would be a kind of "automatic" exposure compensation. It would only really work for the evaluative mode where the camera would evaluate 21 areas and pick the one to expose on.

For digicams that offer a live histogram, the same concept could be used to calculate exposure, only instead on relying on 21 metering zones, the camera could use the actual histogram and calculate exposure such that the histogram is 'exposed to the right'.

The idea has been floating around for quite a while. And I'm surprised that no company, as far as I know, has implemented it in some fashion. Of course there would have to be a way of adjusting a threshold of some sort. Some highlights you don't care about blowing. A spot of sunlight reflecting off of the chrome trim of an automobile, or sunlight on water can and should be blown, otherwise you'll have nothing but noise when you're adjusting the exposure later. But it should be possible to do. And since highlights are the key to optimum digital exposure, the ideal metering system should be looking at highlights.

I think Peter has hit the nail on the head with the issue of the "threshold". The difficulty would be that the variability of that highlight you are trying to save detail in may be 5 or 10 stops.

One example I've seen is in the impementation of the Canon ETTL-II flash system. It is sensitive to reflective items in the photo, cutting flash output to try and prevent highlight details from blowing out. You can see this by taking a picture in a mirror. If the flash head is visible in the shot, the system intervenes drastically cutting output by 3-4 stops to try to save the detail in the flash head.

Using your example of a shooting dancers in a theatre, the camera would have to make an intelligent choice about whether you were trying to capture highlight detail in the dancer's costume (at 3 stops over) or if you were trying to capture detail in the spotlight you accidentally included in the frame (at 20 stops over). There may not be a consistent threshold to use, and the incorrect choice could give you an unuseable shot.

To me, the real problem is not that the highlights clip, it's that they look so bad when they do so!

The irony is that in a digital sensor, you have, in effect, a -REAL- accurate (and real expensive) light measuring device since the whole sensor is a photon counter...

The way I see it, a half push of the shutter release could lock onto both auto-focus while at the same time shooting a non-recorded sensor capture that would evaluate every pixel in the capture and choose the optimal real capture exposure. A function button could lock the exposure setting until you hit the button again to release and re-meter...

This would be particularly useful to have the camera TELL YOU the scene dynamic range (assuming the camera makers would finally get real and adopt a standard). Then you could dial in a compensation factor based upon whether you want to bias the exposure to the shadows or the highlights-in the case where the scene is beyond the dynamic range.

Unfortunately, camera makers still seem to want to put in a separate light measuring device-a meter-when the real light measuring device should be the sensor itself.

Just a thought...

Jeff's suggestion makes a lot of sense, but of course it has not been implemented.

However, some progress is being made. I'm not familiar with Canon metering, but the Nikons have a CCD for metering that contains 1005 cells (alternating RGB) arranged in 15 rows and 67 columns covering virtually the entire frame. This sampling is not as extensive as Jeff suggests, but can be done in real time without pre-exposure. It should give the camera a good indication of the dynamic range and color content of the scene and probably could be put to better use for pre-exposure evaluation by the user and in evaluative metering.

One constant complaint of D70 users was "underexposure"--i.e. the camera refused to expose to the right in a high dynamic range scene. Apologists said this was to protect the highlights, but it made absolutely no sense to have images with the brightest stop of the histogram completely unpopulated. It was quite frustrating for advocates of ETTR. I suspect the exposure algorithm was adapted from their film cameras and not really appropriate for digital. However, the D200 in its matrix evaluative mode is much better--one has to override the camera much less often in order to expose to the right. For once, Nikon has listened to its users.

Live histograms are likely not that far off (my recollection is that it's already implemented on at least 1 dslr).

I suspect it would be fairly easy to dial in a % permissible clipping, either highlights or shadows.

As an interim step, I'd like more accurate RAW based histograms (even if only after the shot). I often find that my brackets are not required and even if the histogram showed clipping at the time of the shot in the RAW image there is actually no clipping.

It can help to know your sensor range. My 1DsMII for example goes up to + 2 2/3 stop and can recover 1 more stop (most of the time) in ACR making my usable range to + 3 2/3 of a given tone spot metered to normal exposure.

I carry a spot meter because of the inaccuracy of the histogram. In many shot it won't matter much, but in high contrast scenes you just need the last bit of usable range you can get to expose correctly to the right, to have nice clean shadows.

I agree with Tim Grey (above) that a raw based histogram would be useful, and it should not be white balanced. Sergio is taking great pains to expose to the right, but with both Canon and Nikon digital SRLs with daylight exposure, the output of the red and blue channels is considerably (about 1 stop) less than the green, and this is not shown on the camera histogram. During conversion, a multiplier is applied to the blue and red channels to equalize them with the green.

The green channel histogram may be exposed properly to the right, but the blue and red fall short of the right. One way to balance the channels and gain better dynamic range is to use a cc100M filter over the lens to hold back the green light and equalize the channels. One then performs white balance with the filter in place and the blue and red multipliers then approach unity. One can gain up to 1 stop of DR with this technique. Here is an example of an unfiltered daylight exposure for the D200:

Click to view attachment

For some images showing this effect with the D200 and EOS 1D Mark II see this thread in the Adobe Camera Raw forum:

http://www.adobeforums.com/cgi-bin/webx?14...85c.3bbf0e1d/10

I originally learned of this technique from a post on the DPReview forum by Julia Borg.

in high contrast situations you can easily hold the highlights from clipping while horribly underexposing your subject. Backlighting would be a good example and probably pretty relevent to your stage shooting. If you have a scene with a lot of white the camera would hold the whites and but the midtones would underexpose. The trick with digital is to decide where you want the highlights, sometimes you have to blow highlights for a good exposure. An ETTR metering mode would cause underexposure in too many situations.

There just isn't enough DR in the highlights with digital to always capture a high contrast scene in one frame which is why HDR gives such great noise, it uses the ETTR for each level of the brightness so you have the maximum information, not only in the highlights but in the shadows too. ETTR is better for the shadows than not doing so, and far better for low contrast scenes, but it is only part of the technique needed to capture the full range of a scene with good quality. At a certain level of contrast ranges you need more than one exposure or you need to work out and understand what highlights you don't mind sacrificing for cleaner midtones.

You are always going to have a two sensor (metering and capture) solution whilst manufacturers retain the current capture sensor design. It is just not feasible to have a live capture type metering with, say, Canon's CMOS sensor design. This is part of the reason for having the two sensors separate. I also suspect that the metering sensor has a greater dynamic range compared with the capture sensor to provide the 0-20ev range that is required to accurately meter a scene. If you do go for an all in one design then the capture sensor would need to have a 0ev to 20 ev dynamic range which is not yet economically realistic, but in 2-5 years may be the panacea that all photographers are looking for.

This doesn't preclude, though, metering sensors with larger numbers of pixels (like Nikon's) and the ability through that type of sensor to provide a live histogram, preferably as an overlay in the viewfinder. However, this may ultimately come down to market demand balanced up against the costs of implementing such a solution.

My point and squirt (Panasonic FX-01) provides live histogram, which is useful given its propensity to clip highlights - though as reviews have pointed out the dynamic range is not great and the sensor is noisier than most would prefer. Having a similar feature on my DSLR would be a welcome addition to existing metering modes - though in reality I can cope quite well with the tools I have at the moment provided a certain amount of learning experience is factored into adjustments and RAW is used to provide additional latitude in the post processing.

dtrayers, excellent suggestion. Why don't you contact Chuck Westfall of Canon USA about it. He seems to be their ear to the ground on user suggestions. He writes a monthy article/response to users'questions and suggestions, and is generally a good source of information if you are having problems with something Canon.

http://dirckhalstead.org/issue0604/westfall.html

At the bottom of the article there is a link to his e-mail.

I think the answer is what the answer has always been. The photographer needs to learn how to expose correctly for the equipment and the situation. Machines cannot make subjective decisions which is what good exposures are based on in complex situations. No silver bullet here, I am afraid.

I am not sure a RAW histogram has any practical value as it would be difficult to evaluate unless you convert it first (which is what RAW processing previews do anyway). Also displaying RAW data is not possible either - which is why a thumbnail file must be created so you can preview the image on the camera. Even if you could display the RAW data, it would be hard to evaluate the image as it is not easy to equate it to the converted data (and forget about white balance).

A RAW histogram would be trivially simple to implement. You wouldn't need to do any Bayer interpolation, white balance adjustments, or any color processing whatsoever. Simply create a bar graph with 32 segments from left to right (32 vertical bars going left to right) with each bar representing 1/3 stop of exposure. The rightmost bar should be red, and its height should represent the number of clipped pixels in the RAW data. The next two bars should be yellow, and their heights represent pixels that are within 1/3 stop of clipping, and between 2/3 and 1/3 of a stop from the clip value, respectively. The remaining bars should represent successively decreasing 1/3-stop exposure intervals, with the leftmost bar or three colored yellow to indicate the possibility of increased noise levels in those exposure levels. A simple lookup table in the camera firmware would suffice to indicate which bar any given RAW value should be assigned, to make the 1/3 stop per bar paradigm work properly. This would require far less firmware programming than the current converted RAW-to-JPEG bastardgram, work perfectly with any sensor regardless of whether it was Bayer-pattern or monochrome, and indicate to the user the exact exposure adjustment necessary to achieve ideal exposure with a single "polaroid" exposure test shot. What's not to like?

I was thinking along the same lines. What Jonathin says is quite true, but I'm not certain how useful raw histograms would be.

On the Adobe Camera Raw forum I once suggested that a raw histogram would be useful, but Bruce Fraser said that it would not be very helpful because, with linear data, all of the data are to the left of the histogram. I wasn't convinced until I did some testing.

Here is a shot of a red flower in which the red channel appears blown as shown in the ACR preview with sRGB:

Click to view attachment

However, with ProPhotoRGB the red channel is no longer blown:

Click to view attachment

Here is the raw histogram from the DCRaw conversion:

Click to view attachment

The raw histogram shows that no channel is blown, but is indeed hard to interpret just like Bruce said it would be. Comments are welcome.

Even if we keep the current metering devices of the cameras, it would still be highly valuable to have some form of automatic expose to the right double exposure:

- first exposure is done based on camera metering device, in a way as to fully capture the non specular highlights,
- second exposure is over-exposed enough compared to the first one so as to get non clipped non specular highlights (based on true RAW histogram data) while fully using the available DR.

The problem is that

- companies like Canon and Nikon just won't release half cooked cumbersome functions, even if they were potentially very useful to knowledgeable users...
- the resulting images would be over-exposed, and would require the typical expossed the right images post-treatment -> many users would complain that the thing doesn't work well,
- ...

Cheers,
Bernard

What is hard about it? The last half is one stop, the quarter to the left of that is another stop; the eight to the left of that is another; the histogram could have dark lines for stops, and thinner lines for 1/3 stops.

That shot should have been taken with 1.3 stops more exposure with almost no clipping, and 2 stops with light clipping, even if it took a higher ISO to do so.

The problem, however, is that most RAW converters don't understand exposing to the right. Setting the Exposure slider in ACR to -2, for example, does not necessarily divide the RAW data by 4, as you'd expect. It may do that to the midtones, but the highlights might be divided by quite less, distorting the transfer curve of RAW to RGB.

Somehow I get an impression you guys are talking about the exposure to the LEFT. "Just make sure highlights are not blown (an unrealistic demand in most cases, especially in the low light situations) and push everything else to the left".

Exposing to the right is supposed to take advantage of the fact that most of the bits in a RAW file are dedicated to the highlights... Which is unfortunate cause sensors don't really capture that many details in highlights in the first place, no matter how many bits are dedicated to them.

The idea of ETTR is to squeese as many details as possible into shadows by moving everything else to the right as far as possible, not to push the shadows out of the picture based on highlights (by moving everything else to the left).

C1 does this quite well, at least with the 1 stop EC's I measured.. IIRC I did a similar test with ACR and it also did ok.

- DL

As far as live histogram - you either need two separate sensors or not an SLR. Both of which exist already.

What are you basing that on?

The sensors capture highlight the best, right up to the saturation point at the lowest ISO, or RAW clipping point if that is lower, and at the higher ISOs. Perhaps you are defining highlights as the areas that sometimes clip? The whole point is getting the highewst exposure possible without clipping desired highlights. Clipping them is not "exposing to the right" properly.

I've found that I can fudge an expose-to-the-right mode by selecting Aperture priority on my Rebel, and then stepping up my exposure compensation (by a fraction of a stop or a full stop, depending on the shutter speed that the camera decides and by how close it is to where I want it). I've found that my Rebel on Ap. Prior. mode generally makes a reasonable exposure in regards to preventing clipping on highlights (with exp. comp. at baseline), but I've found that the little bit of exposure compensation makes it JUST right for my purposes.

If sensors were the best at capturing highlights up to the clipping point you'd be able to pull almost infinite high quality details out of the highlights just before they clip. I'm not an engeneer but it is definitely not what I can observe. The bits are there, the details are not.

"Just don't blow the highlights" exposure is really opposite to trying to shift everything to the right. "Desired" highlights is the key word here - you decide what to clip. I mean - if your image has a narrow dynamic range then of course just overexpose until highlights are about to blow, it should be technically possible at some point in the future to do it automatically... But that's not the case in a situation like described in the original post - it's not a narrow dynamic range.

Almost infinite? I think you mean, 'the maximum detail the system can deliver in the shooting circumstances', don't you?

As I understand, the response of digital imagers is quite different to film in the sense that highlights in film undergo significant compression before total clipping, commonly known as a 'shoulder'.

Digital sensors have a much narrower shoulder. Within half a stop or so, it seems, you can go from a situation of full, uncompressed detail in the highlights to totally blown highlights. There's a much sharper cut-off which presents a major problem for ETTR. It's clearly better to be a 1/2 stop under the correct exposure for ETTR than a 1/2 stop over, if preserving those highlight details is important.

However, what I've just written is an oversimplification (how could it be otherwise. I'm not even sure I know what I'm talking about ). There's another issue relevant here, which is addressed in another current thread, 'expanding dynamic range'. It is unlikely that all 3 channels in a digital sensor are going to 'blow out' at at the same point. The red channel might blow out first, followed by the blue channel, leaving the green channel as pure luminance. It seems there is no way around this, other than to use the right type of filter in front of the lens and do a 'custom WB' before taking the shot.

Now, just how precise do you want to be in your photography?

I was objecting to a description of contemporary sensors as exceptionally adapt at capturing extreme highlight details. Personally I think the sensors are not good at that whatsoever. So you seem to agree with me.

Still - the point of ETTR is overexposing as much as possible so you get the most bits possible dedicated to your image information. Not to measure the extreme highlights and push everything else to the left.

This is a misrepresentation of the technique as described here on the LL and on other sites. I can see why you object to following a technique which is obviously in error, but it isn't ETTR you're describing.

It's not about detail, it's about sensor noise and artifacts, and you definitely shouldn't blow the highlights*:



If reading Michael's article isn't enough, maybe this article by Roger Cavanagh can help.

*There are few rules without exceptions, and I'd just like to point out that there are some highlights that you might desire to blow, such as specular highlights.

There's no doubt that ETTR is the right thing to do for the best quality image. The problem is getting it right. Backing off till the 'blown highlight' warning stops flashing can result in underexposure in my experience. However, this result might be due to RAW converters getting better. I recall when I first started using the D60, I could recover about 2/3rds of a stop in BreezeBrowser. With the latest versions of ACR it seems to be about one and 2/3rds stops.

Consider the following image of a sunrise taken about 4 years ago on one of the rare occasions that I arose before the sun did. It's being converted into a very wide color space, ProPhoto RGB. Brightness and contrast have been taken to a minimum yet it looks as though I have seriously blown the red and green channels.

Click to view attachment

However, if I apply minus 1.5 EC, the apparently blown red and green channels are clearly not blown, as can be seen in the image below.

Click to view attachment


But what about the centre of the sun? That white spot is bigger than a mere specral highlight. Whatever the setting in ACR, it's 255,255,255, even with -4EC, which is okay by me. I'd expect the centre of the sun to be a blown highlight, but I was curious as to what a linear conversion would reveal and was very surprised to find that even that centre white spot does not seem to be blown. It's just a neutral white. The image appears to be actually underexposed by about 1/4 of a stop. That's close enough for me .

Click to view attachment

Well - reread the article. Pay extra attention to the rationale for the technique. It's about bits dedicated to the image data.

It's an overexposure technique geared towards heavy post-processing that benefits from extra bits dedicated to image information, not a guide to correct exposure.

"Don't blow highlights" is a disclaimer, so to speak. "We all know (or at least should by now)" that it's only realistic on narrow dynamic range images. Some highlights are born to be blown (as you've mentioned in your note) .

Quote:

"...Now of course when you look at the RAW file in your favourite RAW processing software, like Camera RAW, the image will likely appear to be too light. That's OK. Just use the available sliders to change the brightness level and contrast so that the data is spread out appropriately and the image looks "right"..."

http://www.luminous-landscape.com/tutorial...ose-right.shtml

Which means "blown" in colloquial English.

You seem to be reading things from me that I didn't write. I said that the range just below clipping of the RAW data is the highest quality recording range.

What are you basing this on?

Every experiment I've ever conducted shows that the upper ranges are the best; the noise is highest there, but the signal is even higher, so the S/N ratio is higher. There is less posterization of your upper tones, as well.

In fact, a super-low-contrast scene exposed with +2 to +3 EC at ISO 400 or 800 has less noise and better subject detail than an ISO 100 shot with 0 EC. All the time; every time. No exceptions noted.



It's not the opposite; it's a different paradigm. It's a matter of what it is that you want to put all the way to the right.

I can agree with that. But the objective of not clipping highlights pushes everything else to the left. The objective of exposure to the right demands an intelligent compromise.

And obviously both of these methods are post-processing oriented.

Well perhaps I should have been less colloquial. In the linear conversion, the centre of the sun appears to be a neutral pale grey with values of 214,214,214, indicating that I could have given a fraction of a stop more exposure.

Technically there are no whites, only shades of grey. The palest shade of grey within our 24 bit color system is represented by the numbers 255,255,255. Isn't that correct?

Ray,

So that we can evaluate your findings, what software did you use for the linear conversion, and what settings did you use. Was white balance applied? What do you mean by linear conversion?

Bill,
I used the linear conversion option in BreezeBrowser. WB as shot would have been applied. All other settings at default. Not accurate enough?

I downloaded the trial version of BreezeBrowserPro. Obviously, I don't know the program, but I can not easily get it to display the actual raw file with no white balance, tone curve, or levels applied. My overexposed color checker looks overexposed in the preview but normally exposed in the conversiion.

I can use manual levels in postprocessing, but I am unable to determine how to display the actual data numbers in the raw file. Personally, I prefer DCRaw for this purpose. Have you tried it?

Bill

Oh. Then it's not blown Sorry for my remark.

I obviously assumed that "white" meant all 255s.

I also had to download the trial version because my older version, which I haven't used for years, would not of course support 5D RAW images. Intitially I could not get the preview window to display the linear image as it should appear and often I would just get a black screen, so I assumed there was a minor bug or system incompatibility somewhere. Oddly enough, it appears to be working today as it should. Don't know whether the system just required a reboot or whether a 'ticking and unticking' of the 'tagged' box did it. Whatever, it's now fine. Below is a screen shot of the BB window.

Click to view attachment

What's interesting here is that the image is not nearly as red as my previous screen shot of a linear conversion, shown earlier in the thread, which was captured within Photoshop. I'm guessing here as to the reason. The linear conversion does not have an embedded profile. My working space is ProPhoto RGB, an extremeley wide gamut space. I did not assign a profile when opening the image in PS but for the purpose of posting on the net, did a conversion to sRGB. The redder image represents how the unchanged numbers, without profile assigned, would look in the sRGB space. The yellower image shows how the adjusted numbers look in sRGB. Does that sound right?



I recently did a Google search on DCRaw and even downloaded something. But I couldn't get the program to work and decided it was beyond my expertise and/or probably not worth the hassle of trying to figure it out. I prefer programs with a user-friendly interface .

Could you tell us your option settings for linear conversion?

I am experimetning with

-v -3 -r 1 1 1 1

The problem with this set is, that it's still scaled, you can't tell for sure, if a channel is blown or not. Should I include

-o 0

to prevent converting it into sRGB, and thus leave the RGB values as they are? Could be a good idea.

Just because the conversion is linear, that does not mean it has not been scaled. The raw file is 0..4095, and the converted linear file should be is 0..255. To convert from the 12 bit raw to 8 bit linear output, the raw pixel value is divided by 16 (a 4 bit right shift in integer math).

I find it difficult to believe that values of 255 in ACR with -4EC do not represent blown channels. Since the linear outupt is around 214, I think it is likely that the divisor is closer to 19 than 16. Since the image is white balanced, we know that multipliers are involved and some scaling has been performed. An analagous situation would be to use the output sliders in Photoshop so that maximum output would be at 214 rather than 255--clipping now occurs at 214 instead of 255.

The easiest way to investigate this possibility is to perform a series of bracketed overexposures and see what the maximum value in the channels is when clipping occurs.

At base ISO most digital cameras set the gain so that sensor saturation results in a data number near full scale in the analog to digital converter. For example, with my Nikon D200, sensor saturation results in RGB values of 254, 249 and 253 expressed as 8 bits and 4064, 3984, and 4048 in 12 bits as determined by conversion with DCRaw with demosaicing and conversion to an RGB image. In this case, clipping occurs in the sensor. Full scale of 4095 is not reached at base ISO, but AD overflow with a maximum data numberof 4095 would occur at higher ISOs and the clipping would be in the AD converter at 4095.

The latest version of CDRaw has a -D switch, which causes totally raw output with no scaling. There is no demosaicing and the output is gray scale 0..4095. If you want RGB output, the above switches with -o 0 should do the trick. The output is scaled by a factor of 16 to convert from 0..4095 (12 bit) to 0..65535 for display at 16 bits. The 12 bit output is very dark in Photoshop 15+1 display (0..32768).

Bill,
I'm not qualified to comment, except to say one might argue equally that, 'I find it difficult to believe that BB's default linear conversion is not accurate'.

I simply don't know. I'm mainly interested in practical results and this sunrise seems to be closely enough exposed to the right for me. Fortunately (or unfortunately depending on perspective), the foreground in the shadows is uninteresting, drought scorched grass. There's no need to bring out the shadows in this shot.

And what has "bits dedicated to the image data" got do do with detail?

It only affects detail to the extent that detail is affected by exposure.

I think we have a terminology problem here, and that we're not communicating quite well, but I'm obviously not in possession of the necessary vocabulary to improve on the situation.

Bill,

thanks for your explanations. Meanwhile, I did some testing an reading of your postings in the Adobe Forum. You say, you use the following options:

-m -n

So, in one case, we agree, since -m is equal to -o 0. See here, and testing prooves this true.

But in the other case, I think your conversion is faulty. The one reason is obvious:

-n is equal to -H 1, which means, that clipped channels are filled with shades of pink, which is not usefull when evaluating highlights. As long as you have no clipped channels in you selection - no problemo. But I don't like that.

For the other reason, I have to go further into detail. It's all about the multipliers. If you set the option -v, you'll see a line like this during conversion:
In lack of a documentation from Dave Coffin, I interpret those values as the multipliers of the direct values to the sensor data, seen as a square crop containing four pixels:

R G B G

My emphasis is on direct, so if you are out for the unbiased sensor data, you'll have to use the option set

-r 1 1 1 1

If you use the -i option, some data of the image file is listed. At the bottom, there are two interesting lines:
These are the values from a Raw file of a Konica Minolta A2.

The first line defines the multipliers for the R G B values, which are supposed to balance the colors with daylight. dcraw sets the smallest value to 1 and adjusts the other colors. This multiplier set is used, if you convert without any (relevant) options set. The computed multipliers displayed are the above mentioned:

0.922500 scaled to 1.000000 (G)
2.094750 : 0.922500 = 2.270731 for R and
1.174418 : 0.922500 = 1.273082 for B

In the second line, you find the values for R G B G according to the set WB. It's the same scheme as above:

256.000000 scaled to 1.000000
271 : 256 = 1.058594
751 : 256 = 2.933594 (pretty extreme value, AWB at tungsten)

Exactly those values, you'll see if using the -w option (Use the color balance specified by the camera).

No, you are using the -n option, which gives the following multipliers:

This is a remarkable notation, since usually, the smallest value is set to 1.000000. So even if you type an option like this:

-r 0.5 1 1 1

it's converted to this form:

If you set the smallest multiplier (0.440387) to 1, you'll see that the ratio is the same as with the camera daylight multipliers (edited). But now he sets the channel with the highest multiplier to one.

Anyway, you see, that using the -n option leads to false R G B values in the image, in the meaning of not as they were recorded. If you want all channels exactly as they were recorded by the sensor, you'll have to use the

-r 1 1 1 1

option, which BTW overrides the -n option.

Oh, yes, thanks for this hint. I was confused, since the values shifted considerably due to the sRGB conversion. The -o 0 or -m otpion solved the problem.

To sum it up, I recommend the following set for linear unbiased conversion:

dcraw.exe -3 -m -r 1 1 1 1 {file name}

Any comments?

That may be what the article emphasizes, but there is more to it than that. The signal-to-noise ratio is higher in the upper RAW ranges of a given ISO, including the distracting banding noise that some cameras exhibit in under-exposures.



Actually, there really shouldn't be that much post-prcoessing to handle a high RAW exposure. At least in theory, the exposure slider in a converter merely needs to be moved to the left (not all of them work like this, though). What requires lots of post-processing sometimes is a "normal" exposure (swhat the camera's metering thinks is normal), and trying topreserve the RAW highlights without darkening the image.

All 214/214/214 over an area does indicate clipping; something was clipped somewhere in the process, and then darkened in the final output.

It is still possible that there is another reason for that. Many DLSRs clip at less than 4095 for various reasons. Canon 1DmkII cameras slip in the low 3700s, depending on the ISO; 5D cameras clip at 3962 at every ISO, etc. I've looked at samples of clipped D200 NEFs that had 4095 at ISO 100 in some verticle lines, but a max of 4024 or thereabouts in alternating lines. The Canon 10D clips lower at ISO 100 and increases slightly up to 800, etc.

see following post

Dennis,

Yes, I think those switches are fine for our purpose, but personally I use the -H 1 option (same as -n), since I do not want any clipping. It seems to me, if you want to evaluate the status of the raw file, you do not want any clippping. The default is -H 0.

In testing with a heavily overexposed Macbeth CC, I saw little, if any, difference between -H 1 and -H 0 switches when using multipliers of 1 and raw output (-o 0). The clipped high lights did not fill with pink with the -H 1 option. I superimposed the two conversions in Photoshop and used the difference blending option, and the screen was totally black--no visible difference. Here is the -H 1 conversion converted to 8 bits and downsized. As you can see, the left two neutral patches of the color checker are blown to white.

Click to view attachment

I then converted the same files but to sRGB (-o 1) and "as shot" white balance (-w) with and without the highlight clipping (-H 0 and -H 1). Here there is a big difference. Also, the multipliers are different. I think the attachments are self explanatory.

-H 0 -w -o 1

Click to view attachment

-H 1 -w -o 1

Click to view attachment

Click to view attachment

Exactly what I said in an earlier post. The sun's disc would most likely have been red in the actual scene and it has been clipped to white. I'm glad someone agrees with me.

But if you have some clipping, your color values in the histogram are corrupt due to pink tinting introduced with the -n or -H option activated.

There is no difference, since any -r setting overwrites any -a, -w or -H (and thus -n) settings. That's what I tried to explain: With -a, -w and -H dcraw set the multipliers to a certain value. But if you set the -r option, those switches are useless, since the -r option now sets the multipliers. So any set like

-n -r 1 1 1 1

or

-H 0 -r 1 1 1 1

or

-w -r 1 1 1 1

are redundand, since -r is stronger than the others.

Here, something is obviously wrong. Your cmd screenshot shows, that you processed two different images, not the same with different settings. Again please check the issue with the -r switch in combination with -w or -H.