In favour of RAW, once again

Discussion in 'General Technical Discussion' started by Iliah, Sep 29, 2005.

  1. Iliah

    Iliah

    Jan 29, 2005
    nowhere
    "What I am interested in is the following technical question about a possible defect in the demosaicing of the capture.

    ...the red channel always seems excessively noisy in skies, particularly in comparison to the green. I never have this problem with scans from film. Do you know what causes the problem to be so severe in digital captures?"

    And do you know? :) I guess yes. But how to minimize the problem? The answer may sound most controversial.
     
  2. Chris101

    Chris101

    Feb 2, 2005
    Arizona
    Hi Iliah,

    I had always assumed that this effect was from there being twice as many green sites as there are red sites in a 'normal' Bayer mosaic pattern. So red has to make up twice as much data as the Green channel does. Blue has always shown more noise than green.
     
  3. Iliah

    Iliah

    Jan 29, 2005
    nowhere
    Dear Chris, how is red channel exposed in daylight shots?
     
  4. Illiah,

    I'm lurking here. I don't understand your question. It is not specific enough to formulate an answer. Remember, only a few of us are anywhere near your level of technical competence.

    By the way, my 2nd Margulis book finally came; the first copy was missing 20 pages right in the most interesting part!

    RIch
     
  5. Iliah

    Iliah

    Jan 29, 2005
    nowhere
    Dear Rich,

    That one w/ missing pages was a collectors' item :)

    Please consider these two renditions of the same image:

    http://www.pochtar.com/_DSC1142CameraWB.jpg

    http://www.pochtar.com/_DSC1142NoWB.jpg

    What you see is essentially RAW sensor output, no demosaicing (files are shrunken by half instead, greens averaged), no nothing - except applying WB to the first file and conversion of both files from linear gamma space to sRGB. If you will compare histograms, you will see that both red and blue channels received amplification during application of white balance. They are underexposed in the shot. You will also see that red channel received more amplification then the blue one, as red channel was underexposed stronger.

    Images are from D2X, so the difference is not so dramatic as with other cameras due to analogue stage of pre-conditioning white balance.
     
    Last edited by a moderator: Sep 30, 2005
  6. TOF guy

    TOF guy

    208
    Mar 11, 2005
    Hi Iliah !

    Seems that this brings us back to another thread which you've posted:
    https://www.nikoncafe.com/vforums/showthread.php?t=54208

    To summarize (assuming I've understood correctly the other thread !):

    The D2X WB coefficients have a fairly flat curve when plotted against color temperature. This is the result of analog stage of pre-conditioning WB.

    This feature can be used to minimize noise in a channel which is underexposed.

    To minimize noise, we expose to the right. Just exposing to the right may leave one - or two - channels underexposed. You're going a step further and recommanding when shooting with the D2X to expose all channels to the right. This can be performed by changing the WB.

    What is controversial is that we've been trained to preset WB to get the proper histogram after demosaicing, so that we may properly evaluate the histogram. Following your advice, I would use WB to minimize noise in all channels. True WB itself is obtained by shooting a grey card.

    Wouldn't you say that in the "ideal" camera, one should be able to record the correct WB coefficients of the scene, and * independently * be able to adjust analogue amplification gain R/G B/G ratios to get the best signal/noise ? These gains would be written in the NEF file header as well. Then at demosaicing time one can apply the product of gain * wb coefficients to get to the final image.

    Is this what you're alluding to in this new thread ?

    Thierry
     
    Last edited by a moderator: Mar 15, 2017
  7. Iliah

    Iliah

    Jan 29, 2005
    nowhere
    Dear Thierry,

    Yes, that is the controversy. White balance in camera causes underexposure and noise.

    As a half-measure we need to see honest histogram in camera - all channels before white balance applied. For now we can resort to "native white balance" - that colour temperature where the r/g/b ratio is as close to 1:1:1 as camera allows.

    Ideally we need better matching of "density generation" in channels, as it is done in slide film emulsions. That involves improving both sensors and filters. Analogue gain adjusted independently might be a welcome step forward.
     
  8. Chris101

    Chris101

    Feb 2, 2005
    Arizona
    Hello Iliah and Rich,

    I've extracted the red channel from the images you've posted Iliah, and taken some 300% crops. I had to go to 300% to find any noise - nice camera! Here they are (don't worry, they're pretty small):

    No WB
    [​IMG]

    Corrected WB
    [​IMG]

    The first thing I notice is that the corrected image is redder - lighter in the red channel (and, you said in the blue channel, but I'm just looking at the red channel here.) Everything is shifted to the right, in fact the sky above the trees but below the clouds in the middle of the image goes from 135 in the 'No WB' picture to 195 in the "Corrected WB' picture. This corrects the wrong color of the 'No WB' color picture.

    However I also see something else at 300%. Looking at the red channel 'No WB' picture, I see a lot of noise in the sky, and in the trees. The noise shows up in the histogram as the short single pixel spikes in the lower, right side of the histogram. Once the WB is applied, these spikes are buried in the data pixels, and red channel is considerably less noisy.

    So, what is the lesson here? White balancing can reduce the amount of red channel noise in low key daylight images. Or can we more generally say that [highlight]all images will have less noise by ampifying noisy channels?[/highlight]
     
  9. Iliah

    Iliah

    Jan 29, 2005
    nowhere
    Dear Chris,

    Noise visibility is different, but the level of noise is exactly the same by measure - I checked. Also, sharpness of white-balanced image is also slightly less in both red and blue channels - by perception and measure.

    If the image would be taken not with camera set to actual colour temperature, but to lower colour temperature - exposure judged by histogram would be longer (as in that tomato thread Thierry referred to). Thus noise in red channel will be diminished.
     
  10. Chris101

    Chris101

    Feb 2, 2005
    Arizona
    I saw the lower acuity too Iliah. I had mentined it in my reply, but then edited it out, because it's so close to the limen, I thought it might be just me.

    On your second point, if you are saying that the in-camera white balance setting will affect the histogram you see on the back of the camera, then yes - it will help you get a better exposure.

    Now my 'but ...' depends on what the 'white balance precondiioning' of the D2x consists of. If it changes the white balance prior to writing a NEF file (the closest any of us will get to a true RAW image) then my previous tests don't apply to that camera. However I found that setting the white balance in camera did not affect the image.

    Your test, shown above, implies that it does. Is this just a D2x thing - what exactly is 'white balance preconditioning'?
     
  11. Holy cow, this is actually beginning to make sense.

    Iliah, you should post more often. :biggrin: :wink:
     
  12. Iliah

    Iliah

    Jan 29, 2005
    nowhere
    Dear Chris,

    White balance pre-conditioning is analogue manipulation of R and B channels in order to keep R:G:B ratio lower. It is shown on the chart we discussed earlier:

    [​IMG]

    White balance per se does not affect exposure, but it affects histogram. So, you do not see on the histogram how data is in fact captured.

    I got some questions re colour, so here is the same white balanced image, but with camera profile applied (no other correction):

    http://www.pochtar.com/_DSC1142CameraBalancePrf.jpg
     
    Last edited by a moderator: Sep 30, 2005
  13. nfoto

    nfoto Guest

    So, does this mean we should strive for an 1.0 ratio and thus set D2X w/b to around 5300 K?

    With the first D1, w/b setting of the camera changed the RAW data. Then, w/b really mattered. I'm not convinced it does today, but keep an open mind on the subject.
     
  14. Iliah

    Iliah

    Jan 29, 2005
    nowhere
    Dear Bjørn,

    Yes, I think 5300 makes sense for D2X. At least I'm getting better results that way, especially at higher ISO. The key thing for me is to distort white balance in a way that judging exposure by histogram I try to reach the exposure where all three channels start touching the right wall.

    With D1 I use colour correction filters whenever it is possible and shoot at 5500K.
     
    Last edited by a moderator: Sep 30, 2005
  15. So is it possible the gist of this is this?

    - shoot at 5300K

    - adjust WB in post
     
  16. Iliah

    Iliah

    Jan 29, 2005
    nowhere
    Dear Ed,

    Yes, that is what I'm doing when I have a scene with regular mix of colours. If reddish/orange/yellowish colours are dominant, I set white balance higher (7000K); for blueish I set it lower (4000K).
     
  17. Many many thanks for those last 2 posts. I've been following the discussion with about as much success as a chimp trying to understand the workings of a nuclear reactor. But I'm quite happy to allow finer minds to work it out and leave me with some practical advice to try out.
     
  18. Chris101

    Chris101

    Feb 2, 2005
    Arizona
    I'm scratching my head. I get the 5300 part - this is where the D2x sensor/filter will give equal illumination across the spectrum. (Should one use 4000-4200K for other Nikon cameras?) This will give the internal jpeg engine the maximum amount of information, and thus display the most accurate histogram on the back of the camera from which to judge exposure. (Right?)

    Now the part that confuses me is the adjustment for subjects which are not neutrally colored. Are the WB shifts (4000 for blue, 7000 for red) just arbitrary? And we adjust for the reflective color of the subject and not the color of the illumination, right?

    ---------

    [:wink:] Mike, compared to white balance, it's simple. A tree in the rainforest is filled with a critical number of chimpanzees. Each chimp has two bananas, one in each hand. When a banana flies into the mouth of a chimp, the chimp jumps. When one chimp lands on another's back, the landee squeezes his bananas and they go flying.

    After the initial banana, the monkey goes flying, lands on another's back, and two more bananas are fed to two other monkeys - pretty soon the whole tree is filled with jumping monkeys and flying bananas. (I don't know how the chimps turned into monkeys.)

    The point is, that if the White Balance is not set correctly, the bananas will look like leaves, and fall out of the tree cause they don't get eaten. The chimps stop jumping and the reaction cools. This happens all the time in nature. Look closely at Iliah's latest picture:
    [​IMG][/:wink:]
     
  19. JeffKohn

    JeffKohn

    Apr 21, 2005
    Houston, TX
    This last part has me a bit confused. Unless I'm totally misunderstanding you I thought there was a thread last week where you recommended using a lower WB for strong reds, not a higher one?
     
  20. Iliah

    Iliah

    Jan 29, 2005
    nowhere
    Dear Jeff,

    Thank you for your remark.

    MY bad, sorry. Settings I suggested above are in reverse. Should read "If reddish/orange/yellowish colours are dominant, I set white balance lower (4000K); for blueish I set it higher (7000K)". Idea is to get the opposite channel to receive full exposure, as it contains details necessary to overcome flatness of the the basically monochromatic subject.
     
Loading...
Similar Threads Forum Date
Jpg vs RAW, interesting look General Technical Discussion Feb 4, 2016
Unbiased RAW vs jpeg blog General Technical Discussion Oct 2, 2015
RAW SETTINGS CS6 General Technical Discussion Jan 5, 2015
Raw File Conversion General Technical Discussion Jul 26, 2014
Is this true? Once again ... on FF vs DX General Technical Discussion Oct 25, 2008