Universal white balance and ETTR

[torger]

ETTR is not exactly a scientifically defined term, so I guess its meaning can differ a bit depending on who explains it.

I can realize that it may be confusing especially for us landscape photographers since a large number of outdoor scenes have some impossible highlight, lamp, sun, star, bright water reflection etc, and then it is better to talk about suitable exposure to get the right look of highlights, a little bit of "glow" around a lamp may be suitable for example. I rarely find it valuable to capture highlights that are too bright for the eye to see when standing at the scene. The moon is an interesting example, for that I usually find it suitable to overexpose it a little, still keep some structure on it but get the glow around it, it often gives the most natural look.

Another problem is that histograms in cameras are so poorly detailed that you won't really detect in them if small highlights are overexposed. Often clipping "blinkies" are only on luminance and will thus often not blink if one channel is clipped etc. It would be very easy for the manufacturers to provide the tools needed to be able to exactly evaluate the exposure, but by some unknown reason they don't think photographers are interested in having those tools. Uniwb seems to improve this a little bit, but I understand those that think it is too little to be worth it.

I think the best way to explain how to expose digitally is that you should concentrate on the highlights, try to get them as close as saturation as possible in order to capture as many photons from the scene as possible for best signal-to-noise ratio in the whole picture, and if not possible to fit them without overexposure then decide what a suitable amount of overexposure is, but keep in mind what is lost is lost so do not overdo it, and that poor camera exposure evaluation tools can fool you, so play it safe and bracket if there is a difficult situation. Sounds pretty much as expose-to-the-right to me.

[RFPhotography]

To be perfectly clear, h I didn't refer to myself as an expert. 

If you actually think about it, the two go hand in hand.  If a scene has specular highlights then it's not going to fit within the range of the sensor because those highlights are going to be several stops brighter than 'regular' light.  It's often even difficult to control specular highlights with HDR or other blending methods because it's sometimes difficult or impractical to bracket that many stops. 

And if you don't know who Jeff is, then I'd suggest doing a bit of research because he knows a hell of a lot more than either of us likely ever will.

[NikoJorj]

I think the best way to explain how to expose digitally is that you should concentrate on the highlights, try to get them as close as saturation as possible in order to capture as many photons from the scene as possible for best signal-to-noise ratio in the whole picture, and if not possible to fit them without overexposure then decide what a suitable amount of overexposure is, but keep in mind what is lost is lost so do not overdo it, and that poor camera exposure evaluation tools can fool you, so play it safe and bracket if there is a difficult situation. Sounds pretty much as expose-to-the-right to me.

Aaaamen, ite missa est, and all that sort of things as far as I am concerned.

It's often even difficult to control specular highlights with HDR or other blending methods because it's sometimes difficult or impractical to bracket that many stops. 

And moreover, those specular haighlight themselves are often best rendered as pure white in a print, so clipping them is not that a problem. The trickier part is to render the "glow" around it, it's where having enough non-clipped data can make a huge difference.
In some cases as the sun veiled by thin clouds, it's not that easy to render and another exposure for the glow itself might be what is needed...

[hjulenissen]

If you actually think about it, the two go hand in hand.  If a scene has specular highlights then it's not going to fit within the range of the sensor because those highlights are going to be several stops brighter than 'regular' light.  It's often even difficult to control specular highlights with HDR or other blending methods because it's sometimes difficult or impractical to bracket that many stops.  

I dont see how any of that contradicts my initial statement. You seem to presuppose that all highlights must always be captured accurately. My point is that sometimes it makes sense to clip them. Sure the image wont fit within the range of the sensor. So what, if the image looks better as a result?

And if you don't know who Jeff is, then I'd suggest doing a bit of research because he knows a hell of a lot more than either of us likely ever will.

I have seen good posts from Schewe. I try to evaluate statements based on their content, not so much on the author. On the bright side, that means that I am open for good arguments from you even though I don't know you.

-h

[RFPhotography]

No, I don't presuppose that all highlights must be captured properly.  I understand that sometimes some highlights may be clipped.  What I'm saying though is that if highlights are clipped then ETTR doesn't apply.  'Better' is a subjective term.  What's better to you may not be to me and vice versa.

[hjulenissen]

No, I don't presuppose that all highlights must be captured properly.  I understand that sometimes some highlights may be clipped.  What I'm saying though is that if highlights are clipped then ETTR doesn't apply.  'Better' is a subjective term.  What's better to you may not be to me and vice versa.

Then we just have do agree that we disagree. I believe that you think that ETTR is about increasing exposure until the brightest pixels nearly saturate. I believe that ETTR is about increasing exposure until the brightest pixels that one care about nearly saturate. In all fairness, that difference is fairly minor in the big picture.

-h

[RFPhotography]

In all fairness, that difference is fairly minor in the big picture.

-h

True.

[Guillermo Luijk]

For those new to uniwb, here's a link (for Canon cameras) http://www.guillermoluijk.com/tutorial/uniwb/index_en.htm

The link is not exclusively for Canon cameras, it's for any camera providing custom WB on a neutral gray card.

Regarding the ETTR discusion, I think it's all about semantics. Anyway IMO saying that ETTR is only of application when absolutely no highlights are clipped, sounds like trying to put a silly restriction on a still 100% valid technique.

ETTR is a concept, not a definition. In a case with specular highlights, one could ETTR the rest of the scene. In a scene of a bride and groom in the shadow with a bright sky in the frame, one could ETTR the sky (so there would be no clipped areas in the capture), but one could prefer to ETTR the bride's dress instead (the highlights of interest), in order to have less noise in the groom's suit (in that case the sky would get clipped, no problem with that regarding the ETTR concept):

Regards

[RFPhotography]

(in that case the sky would get clipped, no problem with that regarding the ETTR concept):

That's a matter of opinion.

[Guillermo Luijk]

That's a matter of opinion.

It is not an opinion, it is a fact that the bride's dress was ETTR.

[RFPhotography]

You missed the point entirely.

[Guillermo Luijk]

You missed the point entirely.

...and that is an opinion.

[madmanchan]

This is why I think improved camera histograms don't address the issue at hand.  I think it would be vastly preferable to have a visualization overlay on the image itself to show which pixels are clipped in the raw data.

As an example, it is common in image processing software to see a visualization of which pixels have clipped (e.g., red highlight warning in ACR).  As you drag Exposure slider to the right, eventually more and more "red warning" pixels show up, right?  Well, would be very nice if you could do this on the camera (with a knob) at capture time ...

[NikoJorj]

This is why I think improved camera histograms don't address the issue at hand.  I think it would be vastly preferable to have a visualization overlay on the image itself to show which pixels are clipped in the raw data.

Rethinking to it (with the almost-ETTR feature of my new EPL1), I'd say you do need both to be efficient in the field : either the photo is underexposed, and then the histogram tells you by how much it is, or it is overexposed and the HL blinker tells you where it is blown.
But if one has to choose (alas, one must with EPL1, no blinkers AND histogram in the same display) I agree the HL blinker is more useful. I still miss the histogram when nothing blinks on the HL side.

[Guillermo Luijk]

This is why I think improved camera histograms don't address the issue at hand.  I think it would be vastly preferable to have a visualization overlay on the image itself to show which pixels are clipped in the raw data.

If I had to choose, I would also prefer visual highlight clipping warning, but not so vastly as you suggest. At least not if there is not also an underexposure warning (certain number of stops below saturation). Otherwise you can be sure not to clip the highlights, but there will remain uncertainty about noise that can be expected when lifting the shadows.

In this sense, a dynamic histogram in EV divisions can even be more accurate and helpful than clip warnings. If this histogram let's you know visually the range of acceptable noise, it can be even used to do ETTL (Expose to the left = expose as low as possible right before having SNR below a threshold in the shadows) in critical shutter and/or DOF applications, where we want to expose as less as possible.

With this kind of histogram:



The RAW shooter just needs to make sure the histogram falls into the range between clipping and noise. Any exposure inside this range will be good enough regarding noise and highlight information.

Regards

[ErikKaffehr]

Hi,

I'd would prefer blinking highlights and also shadows markings. I like the histogram, but it does not show what is clipped. The dynamic histogram suggested by Guillermo is good, but I think that we could simply have a clipping warning.

I really would like to see more live view based viewfinder solutions, and in that context I feel that we should maximize information but minimize viewfinder intrusion.

Best regards
Erik

If I had to choose, I would also prefer highlight clipping, but not so vastly as you suggest. At least not if there is not also an underexposure warning (certain number of stops below saturation). Otherwise you can be sure not to clip the highlights, but there will remain uncertainty about noise that can be expected when lifting the shadows.

In this sense, a dynamic histogram in EV divisions can even be more accurate and helpful than clip warnings. If this histogram let's you know visually the range of acceptable noise, it can be even user to do ETTL (Expose to the left = expose as low as possible right before having SNR below a threshold in the shadows) in critical shutter and/or DOF applications, where we want to expose as less as possible.

With this kind of histogram:



The RAW shooter just needs to make sure the histogram falls into the range between clipping and noise. Any exposure inside this range will be good enough regarding noise and highlight information.

Regards

[RFPhotography]

Just to be clear, you all are talking about a live view image on the back LCD before the image is made, yes?  And it would have to be calculated off some sort of raw (unprocessed light hitting the sensor) data, yes?  Because I think pretty much all cameras have blinking highlight indicators available after capture calculated off JPEG data and the 'knob' in that case is the aperture value, shutter value or exp. comp. and retaking the shot (not always possible).

Or..... one could simply do a reasonable amount of testing with their camera in different conditions so that they know what headroom is available and not have to rely on the histogram or a clipping overlay, yes? 

[fdisilvestro]

Just to be clear, you all are talking about a live view image on the back LCD before the image is made, yes?  And it would have to be calculated off some sort of raw (unprocessed light hitting the sensor) data, yes?  Because I think pretty much all cameras have blinking highlight indicators available after capture calculated off JPEG data and the 'knob' in that case is the aperture value, shutter value or exp. comp. and retaking the shot (not always possible).

Yes, current cameras have that for the rendered JPEG, we would like the information based on the raw values

[madmanchan]

Exactly:  want the clip indicator (overlay, visualization, etc.) to be calculated based on raw image pixel values, before white balance & other color-related processing is applied.

[madmanchan]

Otherwise you can be sure not to clip the highlights, but there will remain uncertainty about noise that can be expected when lifting the shadows.

Guillermo, my thinking is that as long as the data is not clipped, more light is better right?  So just increase the exposure dial till the clip warning overlay appears (over something that is important to you, e.g., can allow some small metallic areas to blow out if you want), and that's the limit.  If this creates an unacceptably long exposure time, then just reduce the exposure time to the acceptable limit.