Ad
Ad
Ad
Pages: « 1 [2] 3 4 »   Bottom of Page
Print
Author Topic: Is expose to the right ever wrong?  (Read 12926 times)
Mark D Segal
Contributor
Sr. Member
*
Offline Offline

Posts: 6976


WWW
« Reply #20 on: December 08, 2011, 01:48:38 PM »
ReplyReply

Possibly, but leaves questions in my mind about the level along the tonal range at which intentionally induced dithering is overcome by the correlation of S:N, and what degree of image enlargement one requires to see any of these effects at work.
Logged

Mark D Segal (formerly MarkDS)
Author: "Scanning Workflows with SilverFast 8....." http://www.luminous-landscape.com/reviews/film/scanning_workflows_with_silverfast_8.shtml
ejmartin
Sr. Member
****
Offline Offline

Posts: 575


« Reply #21 on: December 08, 2011, 02:10:07 PM »
ReplyReply

I think some of the quibble here is over semantics, or rather precision of language.  There is a distinction between 'encoded levels' and 'information'.  Having more levels does not necessarily mean more information.  What photographers who employ ETTR understand intuitively is that by increasing exposure they get smoother tonal gradations, and that is indeed because there is more information in the image due to higher S/N.  There are indeed more distinguishable levels because that number is governed by the S/N; and this number of distinguishable levels, it should be noted, is distinct from the number of encoded levels which is the number of raw values encompassing any given exposure zone in the image.  One can play all sorts of tricks with the latter without changing the S/N and thus the image quality, as Guillermo's example shows.
Logged

emil
Schewe
Sr. Member
****
Offline Offline

Posts: 5500


WWW
« Reply #22 on: December 08, 2011, 02:15:52 PM »
ReplyReply

Unless I'm missing something, which is entirely possible, this seems to set aside the notion that through increased exposure of the darker scene tones, they become lighter captured tones as one is moving them rightward up the scale where there is a higher number of encoded levels, and the S/N is greater.

No...actually it dovetails fine. The bottom line is it's the increase of photons caused by ETTR that increases the SNR. The bit about "bits" or levels or tones originally written about by Mike is what's the confusing issue. ETTR works because it's the increase in the photon count that improves the signal-more photons = better signal-not more bits in the shadows.
Logged
Mark D Segal
Contributor
Sr. Member
*
Offline Offline

Posts: 6976


WWW
« Reply #23 on: December 08, 2011, 02:52:37 PM »
ReplyReply

No...actually it dovetails fine. The bottom line is it's the increase of photons caused by ETTR that increases the SNR. The bit about "bits" or levels or tones originally written about by Mike is what's the confusing issue. ETTR works because it's the increase in the photon count that improves the signal-more photons = better signal-not more bits in the shadows.

OK, now taking this one step further. I have cases "A" and "B". In case "A" I don't care about ETTR, I make an exposure, the scene's dark tones emerge really dark - not many photons - the histogram is positioned relatively leftward. In case "B" I do care about ETTR, so I do it - safely - and those same scene dark tones are hitting the sensor at a higher photon count, because my ETTR is achieved either by a wider lens opening or a lower shutter speed. Absent further intervention they would look "over-exposed".

Now, when I open both images in LR and I want to optimize luminosity (brightness and contrast), in Case A I would find myself lightening up the shadow areas and revealing more of the latent noise in the image. In Case B I would be remapping the brightness of the shadow areas to darker values, but if ETTR is to be a useful recommendation at least with respect to quarter-tones, the Case B process should reveal less noise and perhaps have more refined tonal gradation than the Case A result. ??
Logged

Mark D Segal (formerly MarkDS)
Author: "Scanning Workflows with SilverFast 8....." http://www.luminous-landscape.com/reviews/film/scanning_workflows_with_silverfast_8.shtml
Guillermo Luijk
Sr. Member
****
Offline Offline

Posts: 1291



WWW
« Reply #24 on: December 08, 2011, 03:03:56 PM »
ReplyReply

the Case B process should reveal less noise and perhaps have more refined tonal gradation than the Case A result. ??

The case B will have less noise and a more refined tonal gradation, but the only visible effect you will see is less visible noise. In other words, there is nothing about tonal gradation in Case B that you could be missing in Case A, because in case A the lower SNR dithered any lack of tonal gradation. Both shots A and B have the same robustness against posterization (so none of them is particularly lacking tonal gradation), but A is noisier.

Take a typical isoless camera such as the Pentax K5. Take a shot at ISO100 with 2 stops headroom in the highlights. Now take another shot at ISO400, a perfectly ETTR'ed RAW file. The ISO400 shot has a more refined gradation (the RAW file has 4 times more filled levels), but you will not notice it since SNR is the same on both shots. In brief, having more levels not always means a visible softer tonal gradation. For the same reason, having less levels not always means seeing a poorer tonal gradation (Case A vs Case B).

« Last Edit: December 09, 2011, 04:13:16 AM by Guillermo Luijk » Logged

craigwashburn
Full Member
***
Offline Offline

Posts: 118


« Reply #25 on: December 24, 2011, 11:08:29 AM »
ReplyReply

In real world shooting, flare is a very good reason not to ETTR.  Flare from a bright sky or other light sources will wash out more detail than you save by reducing noise.  It does not take much overexposure for moderate flare to arrive.

Other reasons are motion blur and saving yourself time in post.  There's splitting hairs about S/N, but these are things most people will never encounter, certainly not in a real world viewing and real world shooting.

The above are, however.


Logged
bjanes
Sr. Member
****
Offline Offline

Posts: 2825



« Reply #26 on: December 24, 2011, 11:27:39 AM »
ReplyReply

In real world shooting, flare is a very good reason not to ETTR.  Flare from a bright sky or other light sources will wash out more detail than you save by reducing noise.  It does not take much overexposure for moderate flare to arrive.

I don't think your hypothesis is correct. Veiling flare is proportional to luminance and will be increased when you ETTR, but when you normalize the image by decreasing exposure in the raw converter, the flare will be proportionately reduced.

Regards,

Bill
Logged
ErikKaffehr
Sr. Member
****
Online Online

Posts: 7655


WWW
« Reply #27 on: December 24, 2011, 02:22:37 PM »
ReplyReply

Hi,

In my view, ETTR is always OK if correctly done. On the other hand, a correctly exposed image can be a boring image. We can capture a lot of DR but mapping it down to screen may be very hard.

The image here shows an example of an image straight out of camera and the same image after some manipulation in CS5:



The interpretation may be a bit psychedelic but illustrates how much information is buried in a digital image. The workflow is described here: http://echophoto.dnsalias.net/ekr/index.php/photoarticles/61-hdr-tone-mapping-on-ordinary-image

Best regards
Erik


Luminous Landscape has taught me to err on the side of over-exposure and not under-exposure.
Provided that one does not clip any highlight information that is desired, is it ever wrong to expose to the right?
Are there circumstances where one might get better results not doing ETTR (expose to the right)?
« Last Edit: December 24, 2011, 02:24:22 PM by ErikKaffehr » Logged

craigwashburn
Full Member
***
Offline Offline

Posts: 118


« Reply #28 on: December 25, 2011, 03:44:07 PM »
ReplyReply

I don't think your hypothesis is correct. Veiling flare is proportional to luminance and will be increased when you ETTR, but when you normalize the image by decreasing exposure in the raw converter, the flare will be proportionately reduced.

Regards,

Bill


Not necessarily.  Shoot two series of images, one with a light source surrounded by a light colored background, and one by a dark colored (windows and curtains work nicely)  The dark colored will have more noticeable bleed into it at moderate exposures.

Not sure why this is, though I suspect its because veiling glare, and its causes and solutions, are more complex than a linear relationship.  Perception may play a part as well, which tends to cause havoc when debating merits of things like S/N Smiley

Logged
bjanes
Sr. Member
****
Offline Offline

Posts: 2825



« Reply #29 on: December 25, 2011, 03:53:16 PM »
ReplyReply

Not necessarily.  Shoot two series of images, one with a light source surrounded by a light colored background, and one by a dark colored (windows and curtains work nicely)  The dark colored will have more noticeable bleed into it at moderate exposures.

And what do these conditions have to do with ETTR?

Regards,

Bill
Logged
craigwashburn
Full Member
***
Offline Offline

Posts: 118


« Reply #30 on: December 25, 2011, 04:59:04 PM »
ReplyReply

And what do these conditions have to do with ETTR?

Regards,

Bill

Hi Bill,

ETTR results in more noticeable glare in real world conditions.  Modern cameras are fairly clean noise wise, even at fairly high ISO it's inoffensive, glare is more of a problem in reducing detail than noise is.  Thus, the photographer will want to correct for glare that might be linear (veiling global glare), or non-linear (localized) in post.  This all takes time.

At one time, when noise was a bigger issue, ETTR made sense, if you didn't have highlights to control.  But now, ETTR as a general shooting rule is deprecated and, at best, a waste of time.  Expose properly and be done with it.  Move on to the next shot rather than worrying about whether you've saturated the pixels enough Smiley

I shoot a lot of architecture and glare is a frequent unwelcome guest.  Here's a very interesting paper on (rather extreme) glare and various solutions:  http://graphics.stanford.edu/papers/glare_removal/
Logged
bjanes
Sr. Member
****
Offline Offline

Posts: 2825



« Reply #31 on: December 25, 2011, 05:36:59 PM »
ReplyReply

Hi Bill,

ETTR results in more noticeable glare in real world conditions.  Modern cameras are fairly clean noise wise, even at fairly high ISO it's inoffensive, glare is more of a problem in reducing detail than noise is.  Thus, the photographer will want to correct for glare that might be linear (veiling global glare), or non-linear (localized) in post.  This all takes time.

At one time, when noise was a bigger issue, ETTR made sense, if you didn't have highlights to control.  But now, ETTR as a general shooting rule is deprecated and, at best, a waste of time.  Expose properly and be done with it.  Move on to the next shot rather than worrying about whether you've saturated the pixels enough Smiley

I shoot a lot of architecture and glare is a frequent unwelcome guest.  Here's a very interesting paper on (rather extreme) glare and various solutions:  http://graphics.stanford.edu/papers/glare_removal/

I still see no evidence that you have brought forth that ETTR with proper normalization produces more veiling glare than an expose to the left strategy with no adjustment. The Stanford paper is interesting, but nowhere do they mention ETTL as a solution to this problem. Do you have any examples to support your thesis?

Regards,

Bill

Logged
jeremypayne
Guest
« Reply #32 on: December 25, 2011, 05:45:09 PM »
ReplyReply

ETTR isn't 'deprecated' ... It is the axiomatic principle of digital exposure.
Logged
craigwashburn
Full Member
***
Offline Offline

Posts: 118


« Reply #33 on: December 25, 2011, 06:41:00 PM »
ReplyReply

Hi Bill and Jeremy,

I think you misunderstand my point.  ETTR is a waste of time because its benefits are negligible, and may even be harmful to an image when in real world conditions.  By 'harm' I mean degradation from non-linear glare that normalizing cannot fix, or as simple as motion blur from a longer exposure time.  In laboratory conditions, where you don't have any glare and isolate all vibration... it may result in slightly better results in some pixels (would these be visible on a normal display or print?), but what benefit is this outside of there?

Glare is often non-linear in real-world shooting, which is partly why the deconvolution technique in the Stanford paper didn't work so well and why they came up with the method to deal with it physically.  Normalizing an ETTR exposure will not always fix it.  This is one reason why I consider glare to be a downside to ETTR.  I find this glare to be more offensive than slight noise that is not viewable under real world conditions.




Logged
BartvanderWolf
Sr. Member
****
Online Online

Posts: 3767


« Reply #34 on: December 25, 2011, 07:00:07 PM »
ReplyReply

Hi Bill,

ETTR results in more noticeable glare in real world conditions.  Modern cameras are fairly clean noise wise, even at fairly high ISO it's inoffensive, glare is more of a problem in reducing detail than noise is.  Thus, the photographer will want to correct for glare that might be linear (veiling global glare), or non-linear (localized) in post.  This all takes time.

Hi Craig,

I agree that global veiling glare (very low spatial frequency due to outside-of-FOV) and more localized glare/flare (due to bright light sources, or bright reflective surfaces, within the FOV) can be a pain. However, I'm undecided about it's non-linearity in its contribution to the final exposure. Given an amount of flare for a given scene with a given optical path and exposure level, only increasing the exposure level would result in a linear contribution of all elements that add to the exposure, assuming that the sensor has a linear response (which CCDs an CMOS devices do in general, save for provisions for draining overexposure), or so it would seem.

Yet John McCann also suggests that the veiling glare contribution is non-linear. I can imagine that being the case for intra-occular glare, because the change of the pupil diameter of our eyes is a function of the brightness, and thus of the optical and thus glare performance. However, when only the exposure time is a variable, I sofar fail to see the non-linear contributing component(s).

Quote
I shoot a lot of architecture and glare is a frequent unwelcome guest.  Here's a very interesting paper on (rather extreme) glare and various solutions:  http://graphics.stanford.edu/papers/glare_removal/

Indeed, architecture is a more obvious scenario in which I also encounter the detrimental effects of veiling glare (which can be minimized on a given lens by using a good adjustable lens hood, and in general by using lenses with few lens-groups, good coating, and proper internal lens edge and barrel blackening). Some lenses even use multiple apertures to suppress internal reflections.

The only reason for non-linearity that I can identify sofar, is that it's spatially variant (i.e. subject to local brightness), but I don't see the non-linear aspect of a uniform change in exposure level (by varying exposure time, not by changing aperture (which is a no-no in HDR photography)) for a given scene.

Given John McCann's examples, I feel that I'm overlooking something obvious ..., but sofar I'm unable to pin-point it.

Cheers,
Bart
Logged
craigwashburn
Full Member
***
Offline Offline

Posts: 118


« Reply #35 on: December 25, 2011, 08:26:40 PM »
ReplyReply

Hi Craig,

I agree that global veiling glare (very low spatial frequency due to outside-of-FOV) and more localized glare/flare (due to bright light sources, or bright reflective surfaces, within the FOV) can be a pain. However, I'm undecided about it's non-linearity in its contribution to the final exposure. Given an amount of flare for a given scene with a given optical path and exposure level, only increasing the exposure level would result in a linear contribution of all elements that add to the exposure, assuming that the sensor has a linear response (which CCDs an CMOS devices do in general, save for provisions for draining overexposure), or so it would seem.

Yet John McCann also suggests that the veiling glare contribution is non-linear. I can imagine that being the case for intra-occular glare, because the change of the pupil diameter of our eyes is a function of the brightness, and thus of the optical and thus glare performance. However, when only the exposure time is a variable, I sofar fail to see the non-linear contributing component(s).

Indeed, architecture is a more obvious scenario in which I also encounter the detrimental effects of veiling glare (which can be minimized on a given lens by using a good adjustable lens hood, and in general by using lenses with few lens-groups, good coating, and proper internal lens edge and barrel blackening). Some lenses even use multiple apertures to suppress internal reflections.

The only reason for non-linearity that I can identify sofar, is that it's spatially variant (i.e. subject to local brightness), but I don't see the non-linear aspect of a uniform change in exposure level (by varying exposure time, not by changing aperture (which is a no-no in HDR photography)) for a given scene.

Given John McCann's examples, I feel that I'm overlooking something obvious ..., but sofar I'm unable to pin-point it.

Cheers,
Bart


I am not sure about the non-linear glare either.  A google search on non-linear veiling glare brings back some interesting papers, but they seem to be for very specific situations, such as reducing the reflection of a car dash off a windshield, where geometry makes a difference. 

W/ respect to photography I can think of a couple of issues involving human perception.  Most raw files are processed with a curve applied to them because human vision isn't linear.  This will cause some issues.  And then of course if the flare is saturating a color channel.  One source of glare that is behind the shutter is the sensor itself causing reflections, I would also assume some mirrorbox features and geometry would also contribute.   The Stanford paper mentions this, but doesn't go into detail.  It seems like this would be highly dependent on source position and intensity and some other factors.

In real world shooting, glare tends to be noticeably worse when a source is surrounded by a dark feature rather than a light one.  A bright light against a white ceiling is less problematic that one against a dark.  It may simply be a matter of perception, which is what is really important here, but I bet the default curve applied to raw images contributes.

Logged
bjanes
Sr. Member
****
Offline Offline

Posts: 2825



« Reply #36 on: December 25, 2011, 10:07:33 PM »
ReplyReply

In real world shooting, glare tends to be noticeably worse when a source is surrounded by a dark feature rather than a light one.  A bright light against a white ceiling is less problematic that one against a dark.  It may simply be a matter of perception, which is what is really important here, but I bet the default curve applied to raw images contributes.

I still don't understand how all of this talk about human perception applies to ETTR and veiling glare. Whether or not we ETTR a given scene, the background and other factors do not change and the overall image is the same except for exposure.

Shown below are some shots of a Stouffer wedge with and without masking off glare from the light table. The images are rendered in ACR with a linear tone curve. The image marked 08 has the surrounding brightness of the light table masked off to reduce veiling glare. The image marked 01 Nominal is without masking and is exposed so that step 1 has an sRGB value of  193. Image 03 is exposed to the right so that step 1 has a value of 252. Exposure of image 03 is decreased in ACR so that step one has a value of 193 (the same as image 1 exposed to the left). The shadow values are virtually the same, showing in this case that ETTR does not affect veiling glare.

Regards,

Bill

Logged
hjulenissen
Sr. Member
****
Offline Offline

Posts: 1683


« Reply #37 on: December 26, 2011, 02:31:18 AM »
ReplyReply

or as simple as motion blur from a longer exposure time.
I think that ETTR should be defined as "increase exposure until you have objectional highlight clipping, or objectional motion blur or objectional lack of DOF". In other words: expose hot, but not too hot and not if it makes the image look bad. Then motion blur should not be a problem either?

-h
Logged
Guillermo Luijk
Sr. Member
****
Offline Offline

Posts: 1291



WWW
« Reply #38 on: December 26, 2011, 08:17:26 AM »
ReplyReply

I still don't understand how all of this talk about human perception applies to ETTR and veiling glare. Whether or not we ETTR a given scene, the background and other factors do not change and the overall image is the same except for exposure.

Me neither.

Exposure is about how much time you spend collecting photons. As long as your sensor doesn't get saturated (i.e. as long as you do proper ETTR), the only difference between shots taken at different shutter speeds, once their exposures are matched in post processing, is SNR. Proper ETTR has zero influence in the harmful effects of glare.

Regards
Logged

craigwashburn
Full Member
***
Offline Offline

Posts: 118


« Reply #39 on: December 26, 2011, 09:53:13 AM »
ReplyReply

I still don't understand how all of this talk about human perception applies to ETTR and veiling glare. Whether or not we ETTR a given scene, the background and other factors do not change and the overall image is the same except for exposure.

Shown below are some shots of a Stouffer wedge with and without masking off glare from the light table. The images are rendered in ACR with a linear tone curve. The image marked 08 has the surrounding brightness of the light table masked off to reduce veiling glare. The image marked 01 Nominal is without masking and is exposed so that step 1 has an sRGB value of  193. Image 03 is exposed to the right so that step 1 has a value of 252. Exposure of image 03 is decreased in ACR so that step one has a value of 193 (the same as image 1 exposed to the left). The shadow values are virtually the same, showing in this case that ETTR does not affect veiling glare.

Regards,

Bill



As I said before, in laboratory-like conditions, ETTR may have a modest benefit in the numbers, but in the real world, this bears out differently.   

With your wedge on a light table, you have just about replicated a standard way of measuring global veiling glare in a lens.  McCann does something similar to show that veiling glare limits information available in an image - this actually furthers my point a bit, that ETTR has no benefits and in general shooting, increases chances of a subpar image w/ respect to motion blur and glare.

Logged
Pages: « 1 [2] 3 4 »   Top of Page
Print
Jump to:  

Ad
Ad
Ad