Pages: 1 2 3 [4]   Go Down

Author Topic: DxO Mark  (Read 33347 times)

Panopeeper

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1805
DxO Mark
« Reply #60 on: November 25, 2008, 05:35:22 pm »

Quote from: Tony Beach
These are screenshots from PhotoBola Raw Image Analysis, and they show where (on the bottom) the D300 file is clipped as a result of its being 1/3 stop underexposed compared to the D90 image; 1/250 compared to 1/200.
Tony, obviously you are not prepared to deal with the fact, that measuring the noise on the raw data has nothing common with judging an image based on its visual appearance.

As I posted, the difference in exposure is irrelevant. In fact, noise and DR comparison can be done (and I am often doing that) based on totally unrelated images, i.e. different sceneries, different illuminations, different exposures, etc.

Anyway, I don't know the details of the D2X comparisons, I was referring to my own measurements. However, there is one flaw on their site for sure: they should allow access to the detailed data only after a test verifying that the viewers understand what they are seeing.
Logged
Gabor

ejmartin

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 575
DxO Mark
« Reply #61 on: November 25, 2008, 06:15:57 pm »

Quote from: jani
What's the rationale behind that?

In practice, there is no distinction between SNR=1 and SNR=0; SNR=1 is the threshold of detectability of signal for an isolated pixel, but measuring the noise at zero signal is technically slightly easier.  As I said, though, the two choices yield essentially identical values for the DR.  SNR=1 is the lower threshold for detectability and so it leads to the most optimistic measure of DR; if you like, it is an upper bound for DR.  And as Guillermo said, you can use the SNR plots to determine the DR according to any stricter criterion you may have.
Logged
emil

Guillermo Luijk

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 2005
    • http://www.guillermoluijk.com
DxO Mark
« Reply #62 on: November 25, 2008, 08:25:46 pm »

Quote from: jani
What's the rationale behind that?
pardon? the rationale behind calculating the DR by yourself, or behind chosing 12 dB as threshold?
« Last Edit: November 25, 2008, 08:25:58 pm by GLuijk »
Logged

jani

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1624
    • Øyet
DxO Mark
« Reply #63 on: November 26, 2008, 04:57:10 am »

Quote from: GLuijk
pardon? the rationale behind calculating the DR by yourself, or behind chosing 12 dB as threshold?
The rationale behind chosing 12 dB as the threshold.
Logged
Jan

Guillermo Luijk

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 2005
    • http://www.guillermoluijk.com
DxO Mark
« Reply #64 on: November 26, 2008, 05:18:36 am »

Quote from: jani
The rationale behind chosing 12 dB as the threshold.
I am not an expert but I have found it a good value since when I have seen images (or areas of a given image) with 12 dB of SNR (S/N>=2EV, i.e. S being at least 4 times the std deviation of noise both in RAW levels) I could still distinguish detail, even if quite grainy. Of course this degree of noise will affect more to fine details than to large plain areas, but as a general rule it can be considered a good criteria.

Moreover using that criteria on Emil's plots one gets DR values (between 8-10 EV) that make sense with those figures I obtained time ago less scientifically just by visual inspection on scenes like this:

Canon 350D - ISO100

the 8th f-stop accounting from saturation is signed as -7EV so I estimated a practical usable DR for the photographer of 8 f-stops on this camera.

And later I read on Emil's article that he also suggests this figure (2EV=12 dB) so I consider it a valid tradeoff in photoghraphic applications for several reasons. No idea if the specialised literature could agree with this value.
« Last Edit: November 26, 2008, 05:28:38 am by GLuijk »
Logged

Panopeeper

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1805
DxO Mark
« Reply #65 on: November 26, 2008, 12:08:22 pm »

Quote from: jani
The rationale behind chosing 12 dB as the threshold.
There is nothing rational on that, except that something has to be picked as the basis of comparison.

Expectations vastly differ depending on several parameters, like shooting subjects, presentation forms and sizes, personal preference. Particularly the affected area in the image and its relative lightness has to be taken into account; one tolerates much less noise in the bright areas than in the shadows, and much of that in the shadows gets cut off anyway by black point or by the lower end of the S curve.
Logged
Gabor

james_elliot

  • Jr. Member
  • **
  • Offline Offline
  • Posts: 57
    • http://www.photo-lovers.org
DxO Mark
« Reply #66 on: December 01, 2008, 07:48:04 am »

Has anybody read and has opinions about that article:
http://www.astrosurf.com/buil/nikon_test/test.htm
It claims that the Nikon bodies are applying some kind of raw noise reduction before saving the raw file, and thus reduce also the kind of details that can be captured.
Thus, comparison of noise levels becomes irrelevant.

Cited also on this site:
http://www.photo-lovers.org/fpsensor.html.en
Logged

ejmartin

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 575
DxO Mark
« Reply #67 on: December 01, 2008, 10:29:37 am »

Quote from: james_elliot
Has anybody read and has opinions about that article:
http://www.astrosurf.com/buil/nikon_test/test.htm
It claims that the Nikon bodies are applying some kind of raw noise reduction before saving the raw file, and thus reduce also the kind of details that can be captured.
Thus, comparison of noise levels becomes irrelevant.

Cited also on this site:
http://www.photo-lovers.org/fpsensor.html.en

Nikon does do NR on raw data for exposures of longer than 1/4 or 1 second, depending on the model.  It can't be turned off.  Important for astrophotography, but not important for many other photographic situations.  As far as I am aware this is the only NR that Nikon does to RAW data that can't be turned off by the user.
Logged
emil

Guillermo Luijk

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 2005
    • http://www.guillermoluijk.com
DxO Mark
« Reply #68 on: December 04, 2008, 09:13:53 am »

Quote from: ejmartin
As far as I am aware this is the only NR that Nikon does to RAW data that can't be turned off by the user.
The problem I see here is not actually whether present Nikons do this or not, but the trend of the camera vendors. If they start to play unfair pre-processing their RAW data to obtain cleaner RAW files, it will become difficult to make fair comparisions since detail and sharpness tests should be done in addition to noise measures. And I cannot imagine a standard 'detail' test that can be applied in equal conditions to any camera model (it should play with differences in the optics, in the sensor format,...).
« Last Edit: December 04, 2008, 09:14:02 am by GLuijk »
Logged

madmanchan

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 2115
    • Web
DxO Mark
« Reply #69 on: December 04, 2008, 10:05:13 am »

Quote from: GLuijk
The problem I see here is not actually whether present Nikons do this or not, but the trend of the camera vendors. If they start to play unfair pre-processing their RAW data to obtain cleaner RAW files, it will become difficult to make fair comparisions since detail and sharpness tests should be done in addition to noise measures. ).

Agreed.

But where does one draw the line?

Example 1: Bad pixels (i.e., pixels that are always effectively clipped to white or black). They contribute to standard noise measurements unless you use robust statistics designed to ignore them. Is it ok for a camera to fix bad pixels before writing out the raw file?

Example 2: CFA sensors often have 2 green pixels. Sometimes the two green pixel types differ significantly in response. Is it ok for a camera to balance the greens internally before writing out the raw file? (Keep in mind that green-balancing is effectively a form of noise reduction.)

Example 3: Tiling issues. Large sensors are sometimes composed of multiple pieces which have differing responses. Is it ok for a camera to "balance" the signals across the components before writing out the raw file? (Essentially performing an in-camera calibration.)

Example 4: Lens corrections, such as distortion. Is it ok for a camera to apply geometric distortion correction automatically before writing out the raw file?

(I submit these as real-world examples of things that camera makers either already do, or will be doing.)
« Last Edit: December 04, 2008, 10:07:05 am by madmanchan »
Logged
Eric Chan

Panopeeper

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1805
DxO Mark
« Reply #70 on: December 04, 2008, 11:40:25 am »

Quote from: madmanchan
Example 4: Lens corrections, such as distortion. Is it ok for a camera to apply geometric distortion correction automatically before writing out the raw file?

(I submit these as real-world examples of things that camera makers either already do, or will be doing.)
I doubt that any camera is or will be doing geometric distortion correction; actually, it can not be done at that stage. However, vignetting correction is realistic.
Logged
Gabor

madmanchan

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 2115
    • Web
DxO Mark
« Reply #71 on: December 04, 2008, 12:15:01 pm »

Not long ago, I would have said exactly the same thing.
Logged
Eric Chan

Panopeeper

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1805
DxO Mark
« Reply #72 on: December 04, 2008, 02:07:22 pm »

Quote from: madmanchan
Not long ago, I would have said exactly the same thing.
Well, I am prepared for a surprize, but as of now, I don't see any way to convert a red pixel into a green one in that stage, i.e. without anticipating the WB.
Logged
Gabor

jani

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1624
    • Øyet
DxO Mark
« Reply #73 on: December 05, 2008, 09:21:41 am »

Quote from: GLuijk
(...)

And later I read on Emil's article that he also suggests this figure (2EV=12 dB) so I consider it a valid tradeoff in photoghraphic applications for several reasons. No idea if the specialised literature could agree with this value.
Thanks, to both you and Gabor, I see why it's an approximate, sensible limit.
Logged
Jan
Pages: 1 2 3 [4]   Go Up