Pages: [1]   Go Down

Author Topic: Using HDR For Fine B&W by Christopher Schneiter  (Read 4477 times)

tom b

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1471
    • http://tombrown.id.au
Using HDR For Fine B&W by Christopher Schneiter
« on: July 23, 2011, 01:36:40 am »

I've moved Guillermo Luijk's quote to this new thread so it will have chance of being read than in my earlier post. Hope that this is OK.


"If this had been shot with one capture, either the highlights would have been blown out, or the shadows would have blocked" Christopher said in the article.

What kind of blockage is Christopher talking about? what is a blocked shadow?. After reading the entire article, I see no mention of digital noise, the one and only reason to shoot more than once when doing HDR photography. If shooting once you can preserve the highlights at the same time you get an acceptable level of noise in the shadows, you got it right. There is no point to shoot more.

Bracketing has no magical effects in the final result beyond the degree of visible noise. In fact if bracketing is not needed for noise reasons, it becomes undesirable and tiring (more effort, more time, more memory and CPU resources, ghosting issues,... in brief more limitations).

There is nothing you can get from a bracketed series of pictures, that you cannot get from a single shot as long as it contains detail in the highlights and acceptable noise in the shadows. I'd bet some of the scenes in the article didn't require any bracketing.

There is also no point in bracketing less than 2EV or 3EV apart, the contribution of additional shots is null. Finally, the less noisy sensors become, the less reasons remain for bracketing.
« Last Edit: July 23, 2011, 01:39:27 am by tom b »
Logged
Tom Brown

Bart_van_der_Wolf

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 8911
Re: Using HDR For Fine B&W by Christopher Schneiter
« Reply #1 on: July 24, 2011, 08:05:59 pm »

Bracketing has no magical effects in the final result beyond the degree of visible noise. In fact if bracketing is not needed for noise reasons, it becomes undesirable and tiring (more effort, more time, more memory and CPU resources, ghosting issues,... in brief more limitations).

Hi Tom,

In theory, you're correct. However, in practice images suffer from veiling glare which adds more exposure to the shadows of an individual capture than to its highlights. That's why in exposure blending or HDR exposure bracket compositing, one typically excludes some of the shadow detail from contributing to the composit, not only for noise reasons.

I can recommend some of John McCann's publications (e.g. here, or here, or here if you don't like to read but rather view for an hour) for a more in depth explanation of the issues involved.

Cheers,
Bart
Logged
== If you do what you did, you'll get what you got. ==

tom b

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1471
    • http://tombrown.id.au
Re: Using HDR For Fine B&W by Christopher Schneiter
« Reply #2 on: July 24, 2011, 08:21:45 pm »

Sorry, not my quote but Guillermo Luijk's. He has some very interesting ideas, I wish he would write more in English though.

Cheers,
Logged
Tom Brown

Guillermo Luijk

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 2002
    • http://www.guillermoluijk.com
Re: Using HDR For Fine B&W by Christopher Schneiter
« Reply #3 on: July 25, 2011, 10:51:10 am »

Sorry, not my quote but Guillermo Luijk's. He has some very interesting ideas, I wish he would write more in English though.

Thanks for your appreciation Tom. Unfortunately my native childhood language was Dutch, then I had to learn Spanish and Catalan, and English came in the fourth place. Therefore not too many of my neurons are devoted to this language ;D

Bart, could you please explain (giving some example if possible) why extra shots can be an advantage against veiling glare?.

Regards

Bart_van_der_Wolf

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 8911
Re: Using HDR For Fine B&W by Christopher Schneiter
« Reply #4 on: July 25, 2011, 01:39:07 pm »

Thanks for your appreciation Tom. Unfortunately my native childhood language was Dutch, then I had to learn Spanish and Catalan, and English came in the fourth place. Therefore not too many of my neurons are devoted to this language ;D

Hallo Guillermo,

I already thought your family name could have Dutch roots ...

Quote
Bart, could you please explain (giving some example if possible) why extra shots can be an advantage against veiling glare?.

Well, I can't make it much clearer than John McCann does in the section from 0:07:25 - 0:13:00 in this video. The relevant part is at 0:11:00, but you may need to see his introduction to understand where he gets the data from (4 gray scales, and 3 of them with different neutral density filters, together providing a huge subject dynamic range with linear calibration steps).

In short, the recorded signal from our brackets is not exactly linear in the Raw data, despite the fact that the sensor has a linear response. It is caused by (a variable) amount of veiling glare, which adds non-imaging response mostly to the dark end of the tonescale of each bracketed image. The easy way to reduce some of that non-linear stray light is by clipping the shadow end of each contributing bracket except for the first (longest shadow exposure). However, by clipping part of the tonescale, the dynamic range that each bracket can add to the combined dynamic range is reduced, hence the need for more brackets at smaller intervals.

Now that is the easy way to reduce part of the non-linear contributions after they are created. Better results would be available if we can avoid glare as much as possible to begin with, and when we have a good method to calibrate/linearize each raw data bracket before adding it (with a multiplier) to the combined Dynamic range.

Getting a grip on veiling glare starts with avoiding lens flare as much as possible by using good lenses (few lensgroups, black lens edges, good multi-coating, and internal barrel reflection suppression) with large hoods (larger than usual). With those precautions, we will have less veiling glare, but due to the nature of lenses we will still have a substantial part of the image consisting of veiling glare. That's why it's almost impossible to exceed a real dynamic range per bracket that exceeds 8 or 9 stops, as soon as lenses enter the equation (the dirty little secret that John McCann refers to).

Hope that helps.

Cheers,
Bart
Logged
== If you do what you did, you'll get what you got. ==

kwalsh

  • Full Member
  • ***
  • Offline Offline
  • Posts: 101
Re: Using HDR For Fine B&W by Christopher Schneiter
« Reply #5 on: July 25, 2011, 03:33:06 pm »

Well, I can't make it much clearer than John McCann does in the section from 0:07:25 - 0:13:00 in this video. The relevant part is at 0:11:00, but you may need to see his introduction to understand where he gets the data from (4 gray scales, and 3 of them with different neutral density filters, together providing a huge subject dynamic range with linear calibration steps).

Yes, that shows classic veiling but it doesn't illustrate in any way how additional intermediate exposure steps provide more scene information.  In fact, the two figures show very clearly the veiling glare signature scaling perfectly linearly with exposure as expected.  I think you are getting confused by the figure which shows that the linear step wedge is in fact no longer recorded as a linear step wedge.  That said, as the figure clearly shows, even in the regions corrupted by veiling glare, the exposure relationship of the scene (including glare) from exposure to exposure is perfectly linear.  That is the point of the earlier comment, the intermediate steps in the HDR don't add any scene information.  If the scene can be captured with two widely spaced exposures the intermediate exposures do nothing beyond providing a small increase in total photon count for certain regions and thus reducing photon shot noise.  The effects of veiling glare relative to the scene scale linearly with exposure and won't be improved by a small step size.

Quote
In short, the recorded signal from our brackets is not exactly linear in the Raw data, despite the fact that the sensor has a linear response.

The brackets themselves are perfectly linear relative to each other even if the step wedge itself is corrupted by veiling glare.  I think you are getting confused by the step wedge measurements and the bracket exposures which are not the same thing.  That's the whole point of veiling glare, no longer is a step wedge measured linearly.  Bracketed exposures, however, are perfectly linear for a given spot in the image.  And again, you can see that in the plots in the video - the veiling artifacts scale linearly from exposure to exposure.  Measuring the veiling artifact multiple times at different portions of the sensors linear response curve adds no information.

Quote
It is caused by (a variable) amount of veiling glare, which adds non-imaging response mostly to the dark end of the tonescale of each bracketed image.

No and no.  The veiling glare does not vary with exposure any more than any other flux does.  And it isn't limited to the dark end of the tonescale either.  For the longest exposures the veiling glare has completely eliminated any dark tones from the sensor data.  The flux of veiling glare doesn't change with exposure!  It isn't physically possible for it to do so!  Veiling glare flux is measured exactly like scene flux and its total photon count scales with exposure identically to that of scene flux.  Measuring it a bunch more times in the middle doesn't add any information at all.  That, I believe, was the point of the original post.

Quote
The easy way to reduce some of that non-linear stray light is by clipping the shadow end of each contributing bracket except for the first (longest shadow exposure). However, by clipping part of the tonescale, the dynamic range that each bracket can add to the combined dynamic range is reduced, hence the need for more brackets at smaller intervals.

Since the veiling signature relative to the scene signature is identical in all exposures that do not clip it explain how this would work?

Ken
« Last Edit: July 25, 2011, 08:23:31 pm by kwalsh »
Logged

Guillermo Luijk

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 2002
    • http://www.guillermoluijk.com
Re: Using HDR For Fine B&W by Christopher Schneiter
« Reply #6 on: July 25, 2011, 06:36:57 pm »

Intuitively, I agree with Ken that sensor linearity cannot be altered by any lens or photon count related matters as long as no sensor saturation is taking place. Only when one or more RAW channels get clipped, we can start to think about non-linear behaviour happening.

This shouldn't be a problem in a highlights-preserving single shot scenario, the proposed method for proper HDR capture (as long as noise is acceptable in the shadows to be displayed in the final rendering).

I'll have a look at the video tomorrow.

I have been looking at the video, and I don't think it demonstrates veiling glare means any sensor non-linearity. If the origin of the glare is in the optical domain, it will be linearly recorded by the sensor (basically a photon counter), no matter which exposure is adjusted. Unless clipping occurs, the only difference in the glared areas will be different SNR according to different exposure.

The response curves in the deep shadows are something quite tricky to plot. If you don't work straight in the RAW domain, you will be subject to the somewhat arbitrary black point clipping calculated by your RAW developer. And even working with undemosaiced RAW data, most camera makers clip the black point before encoding the RAW data, making it difficult to properly plot RAW digits vs exposure (clipped noise is not gaussian anymore). So to obtain a perfectly linear plot in the deep shadows is difficult, but this doesn't mean the sensor is behaving non-linearly.

In the highlights area, unless some RAW channel gets clipped sensors are very linear.

Regards
« Last Edit: July 26, 2011, 08:13:05 am by Guillermo Luijk »
Logged

hjulenissen

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 2051
Re: Using HDR For Fine B&W by Christopher Schneiter
« Reply #7 on: July 27, 2011, 03:18:06 pm »

If DR is limited in thhe optical signal path, then I agree that multiple sensor exposures should have limited effect.

But what about using ND-filter to capture the multiple exposures?

-h
Logged

kwalsh

  • Full Member
  • ***
  • Offline Offline
  • Posts: 101
Re: Using HDR For Fine B&W by Christopher Schneiter
« Reply #8 on: July 27, 2011, 03:28:42 pm »

But what about using ND-filter to capture the multiple exposures?

Well if it was just a ND not any improvement, in fact I guess a little degradation because of the extra interfaces to reflect.  You haven't reduced the dynamic range at all, just reduced all the fluxes equally.

I have seen people claim a grad ND for the classic sunset shot does help, which makes sense as you would actually be reducing the dynamic range prior to entering the optics.  I've not seen any controlled demonstration of this, but it does at least make sense in theory for the scenes it would be applicable to.

Ken
Logged

hjulenissen

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 2051
Re: Using HDR For Fine B&W by Christopher Schneiter
« Reply #9 on: July 28, 2011, 05:33:24 am »

Well if it was just a ND not any improvement, in fact I guess a little degradation because of the extra interfaces to reflect.  You haven't reduced the dynamic range at all, just reduced all the fluxes equally.
If the lense behave in a non-linear fashion (?), then reducing the input level globally should change things up.

Please explain the use of the word "flux" in this context.

-h
« Last Edit: July 28, 2011, 05:47:16 am by hjulenissen »
Logged

kwalsh

  • Full Member
  • ***
  • Offline Offline
  • Posts: 101
Re: Using HDR For Fine B&W by Christopher Schneiter
« Reply #10 on: July 28, 2011, 09:32:13 am »

If the lense behave in a non-linear fashion (?)

The lens is actually perfectly linear, even with veiling flare.  We need to be careful to not get confused by the step wedge measurements and what they are really showing.  The graphs in the presentation show that the measurement of the "linear" step wedge has somehow been corrupted by the optics, the graph of the measured patches is no longer linear.  This does not mean that the optics are non-linear however.  What is happening is that the optics are spatially filtering the scene. The lens because of scattered light acts as a very weak low pass filter.  The result is that the light from the brightest patch is spread across the entire image.  This is still a linear process, super-position still applies.  If you reduce the scene brightness with a ND filter the brightest patch still spreads light across the image plane in direct proportion to its intensity.  It corrupts the other patches identically.  The bright patch and its veiling flare are reduced by the ND filter, but all the other patches are reduced by the same amount.

If you put a ND filter in front of the lens nothing about the scattered light or spatial filtering of the lens changes.  You'll get the same corrupted step wedge measurement, you just take a longer exposure to get there.

Quote
Please explain the use of the word "flux" in this context.

Radiant flux (Watts), although for the purposes of this discussion luminous flux would also be equivalent.

http://en.wikipedia.org/wiki/Radiant_flux

In the end the point is optical limits to dynamic range aren't about non-linear optics, the lenses are perfectly linear (the sensor isn't necessarily though, and film is most definitely not).  The issue is the lens is a spatial filter.

Another way to consider it would be a soft focus lens.  Perfectly linear optic, but it won't measure a step wedge as linear.  Put a ND filter on it and you still have a soft focus lens that creates a lot of veiling flare.

Hopefully that was more helpful than confusing!

Ken
Logged

hjulenissen

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 2051
Re: Using HDR For Fine B&W by Christopher Schneiter
« Reply #11 on: July 28, 2011, 12:16:17 pm »

The lens is actually perfectly linear, even with veiling flare. 
Ok. The question then is (besides purchasing really expensive lenses, using large hoods and arranging bright lights cleverly in the scene): can this linear response be removed (sharpening, deconvolution, ...)
Quote
Hopefully that was more helpful than confusing!
I think the explanation was perfectly good.

-h
Logged

kwalsh

  • Full Member
  • ***
  • Offline Offline
  • Posts: 101
Re: Using HDR For Fine B&W by Christopher Schneiter
« Reply #12 on: July 28, 2011, 07:35:11 pm »

Ok. The question then is (besides purchasing really expensive lenses, using large hoods and arranging bright lights cleverly in the scene): can this linear response be removed (sharpening, deconvolution, ...)

Removed, as in reproduce the step wedge with photometric accuracy - seems very unlikely.  The sample images, not unsurprisingly, show a complicated set of flares and reflections that appear dependent on the location of the source in the scene.  Really you'd be trying to remove lens flare, which no one has a good method for.

Could you improve the shadow contrast in post processing to try and reverse some of the effects in a visually pleasing way?  Certainly, to a point, and you'd be amplifying up the noise at the same time.  I've tried that in a few problem images and it works only so far, but it definitely can help.

Deconvolution is notoriously difficult to get to work in the real world and usually only ever works well over a very small kernel (i.e. very small spatial dimensions, as in deconvolution sharpening).  I think it would be a lost cause even on something as well behaved as a soft focus lens, not to mention the problem here with stray light paths.

Anyway, starting to get beyond my direct experience.  I do know from reading on HDR for photometry that sometimes what photographers want - a pretty picture rather than accurate flux measurements - is sometimes disconnected a bit from what the academic literature is often worried about.

Ken

« Last Edit: July 28, 2011, 07:40:09 pm by kwalsh »
Logged
Pages: [1]   Go Up