Pages: 1 [2]   Go Down

Author Topic: Photographic dynamic range, IQ in HDR and the 3 beasts: 5DS R, D810, A7R II  (Read 11430 times)

Jack Hogan

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 798
    • Hikes -more than strolls- with my dog
Re: Photographic dynamic range, IQ in HDR and the 3 beasts: 5DS R, D810, A7R II
« Reply #20 on: December 16, 2015, 10:04:58 am »

I understand Guillermo, I was trying to go perceptual on you :)  What's better looking: X identical images stacked or 3 HDR (-2:0:2 stops)?  I am thinking that the HDR process may be less than ideal, introducing distortions/noise which may be more/less objectionable than a little extra noise in the stacked set?

And thanks Bart, I will try SNS-HDR when I get the chance.
Logged

Guillermo Luijk

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 2005
    • http://www.guillermoluijk.com
Re: Photographic dynamic range, IQ in HDR and the 3 beasts: 5DS R, D810, A7R II
« Reply #21 on: December 16, 2015, 02:34:41 pm »

I understand Guillermo, I was trying to go perceptual on you :)  What's better looking: X identical images stacked or 3 HDR (-2:0:2 stops)?  I am thinking that the HDR process may be less than ideal, introducing distortions/noise which may be more/less objectionable than a little extra noise in the stacked set?

There has to be no difference at all, just an improved SNR in the HDR composite over the stacked version. What you refer to as "better looking" doesn't depend on the source of the information (HDR bracketing vs average stacking), but on the tone mapping applied afterwards as Bart pointed. If the HDR process is performed linearly, there will be no colour or contrast difference between the HDR blend and any of the single shots. So the resulting image will look the same no matter how you started; it will just be less noisy in the HDR.

Real world example: left the least exposed shot preserving the windows, right the HDR composite when information from +2 and +4 shots is added:


HDR fusion map:


100% crops (right is 16 times less noisy):



Just curious: to achieve the same noise reduction (16x) with standard averaging, you would need to average as many as 256 standard shots.

Regards



« Last Edit: December 16, 2015, 03:45:58 pm by Guillermo Luijk »
Logged

bjanes

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 3387
Re: Photographic dynamic range, IQ in HDR and the 3 beasts: 5DS R, D810, A7R II
« Reply #22 on: December 16, 2015, 03:23:25 pm »

A method based on skeptics of human judgement implies it is measuring unjudgeable parameters. More likely it is measuring unpractical or unrealizable difference. One bothering problem to me with some high pixel count sensor is the color shift. This is also mentioned in one of the post in this thread. 

Th problem is the image could show weirded saturation of white balance when sliding the exposure value or fill-n the shade.

Maybe you can give a demonstration with your method to show, channel by channel, the consistency of dynanamic range.

EinstStein,

I see nothing wrong with Guillermo's DR analysis. After all, it closely follows the method outlined by Emil Martinic in this LuLa post and more fully explained in his outstanding and widely cited article on the University of Chicago web site. Emil has a PhD in physics, is a full professor at this prestigious institution, and has published more than a hundred articles in peer reviewed scientific journals. I don't know your qualifications, but imagine that they are less than those of your namesake.

A channel by channel analysis would not be easy to perform, as indicated in the DXO web site here and quoted below.

Full-color sensitivity
The noise on the sensor is processed by a RAW converter to lead to the final image. In particular, RAW converters apply a color rendering which maps the RAW values to RGB values in the final image (for instance, sRGB values). Color rendering includes at least white balancing and a color matrix, as described above. These steps yield an amplification of noise. For each triple RGB in the final image, noise can be predicted from the RAW noise characteristics, the white balance scales, and the color matrix coefficients. The noise on the three color channels is described by a three-dimensional covariance matrix, which is not easy to represent. In the full-color sensitivity tab, we chose to illustrate the noise in the (a,b) plane of the CIE Lab color space. For different values of luminance and different ISO sensitivities, the noise at given colors is represented by an ellipse which shows the colors that are likely to be generated in the image instead of the real color (displayed at the ellipse center). The orientation and the size of the ellipses show the coloration of noise and its amplitude.


Bill
Logged

Jack Hogan

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 798
    • Hikes -more than strolls- with my dog
Re: Photographic dynamic range, IQ in HDR and the 3 beasts: 5DS R, D810, A7R II
« Reply #23 on: December 17, 2015, 03:12:47 pm »

Real world example: left the least exposed shot preserving the windows, right the HDR composite when information from +2 and +4 shots is added:

All right Guillermo, sold.  HDR trumps stacking in that situation.

Jack
Logged

IliasG

  • Newbie
  • *
  • Offline Offline
  • Posts: 22
Re: Photographic dynamic range, IQ in HDR and the 3 beasts: 5DS R, D810, A7R II
« Reply #24 on: December 17, 2015, 05:30:51 pm »


100% crops (right is 16 times less noisy):



Just curious: to achieve the same noise reduction (16x) with standard averaging, you would need to average as many as 256 standard shots.

Regards

Why is the right 16X times less noisy ?. Shooting with 16X more exposure (+4EV) gives sqrt(16) = 4X better SNR .. so I would expect to be equal with 16 standard shots averaged ;)
Logged

Guillermo Luijk

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 2005
    • http://www.guillermoluijk.com
Re: Photographic dynamic range, IQ in HDR and the 3 beasts: 5DS R, D810, A7R II
« Reply #25 on: December 17, 2015, 07:09:58 pm »

Why is the right 16X times less noisy ?. Shooting with 16X more exposure (+4EV) gives sqrt(16) = 4X better SNR .. so I would expect to be equal with 16 standard shots averaged ;)

It is not that simple but luckily it can be estimated more accurately from the SNR curves:
- In the sensor "highlights" where photon noise prevails, SNR increases by 3dB/EV (i.e. SNR is multiplied by sqrt(2) with each extra stop of exposure as you mention) but...
- In the sensor "deep shadows" where read noise prevails, SNR increases by 6dB/EV (i.e. SNR is multiplied by 2 with each extra stop of exposure).

For the case of HDR we are closer to the second situation since we are reallocating the deep shadows of the scene from the sensor's "deep shadows" (many stops below sensor saturation in the least exposed shot that preserves the highlights) to areas 4 stops closer to sensor saturation. That is why I made the 16x times less noise approximation.

But the real resulting SNR improvement will depend on the region we fall into. In my example this is the EV distribution of the +4EV crop:

(RAW histogram in EV)



On average we can assume that the information from the crop is located around 5 stops below sensor saturation, this means in the 0EV shot the same information is on average located 9 stops below sensor saturation. Let's find out the exact SNR improvement from -9 to -5 following the SNR curves of the camera used (DxOMark info):




SNR improvement is around 19,5dB (less than 24dB but more than 12dB), i.e. noise was reduced by 9,44x (less than 16x but more than 4x).

To achieve the same improvement with averaging we would need: 9,44^2=89 standard shots (less than 256 but more than 16).

Since a real world scene is a continuous luminance distribution (my crop spreaded over around 6 stops of DR), the deepest shadows of the scene will always enjoy a larger SNR improvement than the not so deep shadows. This is good news since it means more benefit is obtained where more help is initially needed to produce a well balanced output image in terms of SNR.

Regards
« Last Edit: December 18, 2015, 04:57:46 am by Guillermo Luijk »
Logged

EricV

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 270
Re: Photographic dynamic range, IQ in HDR and the 3 beasts: 5DS R, D810, A7R II
« Reply #26 on: December 18, 2015, 12:27:43 pm »

In the really dark shadows, where sensor read noise is comparable to or higher than photon shot noise, it is better to take one long exposure than many short exposures, in order to incur fewer read noise penalties.  In the less dark shadows, where photon statistics dominate, it should not matter how the photons are collected, as long as the total exposure is the same.

For this same reason, when evaluating sensor noise performance at SNR=1, sensor full-well capacity and read noise are equally important, but when evaluating sensor noise performance at much higher SNR, full-well capacity is almost all that matters.

By definition, dynamic range at SNR=1 is (full-well capacity) / (read noise).  A good approximation to dynamic range at much higher SNR would be (full-well capacity) / (target noise level), where the target noise level is the specified SNR squared.  For example, SNR=16 occurs when the shadows have 256 electrons, so dynamic range at SNR=16 is just (full-well capacity) / 256.
« Last Edit: December 18, 2015, 12:41:13 pm by EricV »
Logged
Pages: 1 [2]   Go Up