No difference between 8 bit and 16 bit. Help?

Discussion in 'Digital Darkroom' started by ron_togger, Feb 25, 2015.

  1. I cannot get my head around this.
    Having exported the same 14 bit NEF Nikon image, firstly as 8bit TIFF, then a 16 bit TIFF (both uncompressed), I applied identical extreme level adjustments to both images. Surprisingly, I cannot see ANY difference in final image quality between the two. Both 8 & 16 bit images look the same, with the same artefacts like gradient banding.
    Logic dictates that raw 14-bit files saved out to 16bit should retain more fine tones compared with a "clipped" 8-bit. Applying photoshop tools like airbrush and grad fills makes an obvious difference - the 16 bit file is much finer, but this only applies to photoshop-rendered objects and not the original photo. It almost seems like the 14bit RAW is in fact only 8 bit.
    Am I missing a major trick here?
    Equipment used:
    Nikon D5100 in RAW mode / View NX2 to convert / Photoshop CS5 to edit. Also used a D3 in RAW mode. Same results.
    Any comments welcome.
     
  2. Since the D5100 only offers 14-bit compressed, it might not be the best camera to use for in this comparison. The D3 should be - provided you started out with the uncompressed RAW or at least lossless compressed. Also, which color space are you making those comparisons in? sRGB might give worse results than using AdobeRGB or ProPhotoRGB - but what you actually see will also depend on the color gamut your monitor can display. And the choice of the monitor will not only affect the colors you see - sometimes banding isn't actually in the image but caused by the display.
    I never made a direct comparison between 8bit and 16bit processing, so I can't offer advice on the "major trick" here.
     
  3. Thanks for the response. It's a puzzle to me. The D3 capture was in Adobe RGB 14/bit uncompressed. Perhaps I am trying the wrong level filters to test these or the differences are too subtle for my iMac monitor...
     
  4. I would imagine this kind of comparison will depend mightily on what the image is. An image filled with all sorts of subtle color information and gradations may show the 8/16 bit difference more clearly than a less complex image. However, I think your observation makes the point that often these differences are not very important or often effectively invisible to observers. Many differences photographers point out are "important" differences between two very similar images are not apparent to objective observers.
     
  5. digitaldog

    digitaldog Andrew Rodney

    http://www.digitalphotopro.com/gear/imaging-tech/the-bit-depth-decision.html
    Keep in mind that compounded editing on 8-bit per color can produce rounding errors that could result in banding. And banding on-screen could be due to the display path and the data could be smooth as can be. When you work with high bit data, you simply don't have to worry that what appears as banding in in your data, you can edit till the cows come home and you'll get no banding on final output. The wider gamut the data, the more likely you'll end up with banding as well.
    Photoshop treats it's data pretty simply; 8-bits pre color or what it calls for simplicity 16-bit data. It might actually be 10-bit, or 14-bit but PS considers this high bit data to be 16-bit.
    Logic dictates that raw 14-bit files saved out to 16bit should retain more fine tones compared with a "clipped" 8-bit.​
    True but you may not see the finer tones or lack thereof without sufficient editing.
    What to see banding? Use the Posterize command in PS. You'll see an excessive application that still happens on your 8-bit per color data. Posterize and enter say 200, then enter 199, 198 etc. You'll eventually see the banding on-screen and it's certainly in the data.
     
  6. sorry... I forgot to add that I deliberately chose images with very light colours and applied extremely heavy level editing to deliberately darken light shades in order to force the banding and other visible artefacts. Some edits DID indeed, produce visibly reduced banding on the 16 bit. I thought I'd solved the puzzle....however...
    ...I then tried one final test: Create an empty 16 bit document, copy and paste the unedited 8 bit image directly into it, keeping the original 16-bit image open as well. I then applied exactly the same processes to both files. Guess what. Both the "original 16-bit" and the up-sampled 8-to-16 bit files behaved exactly the same when processed using heavy filtration. Unlike the first attempt which eventually produced some subtle differences, this did not. I'm completely lost on this one.
     
  7. digitaldog

    digitaldog Andrew Rodney

    ...I then tried one final test: Create an empty 16 bit document, copy and paste the unedited 8 bit image directly into it, keeping the original 16-bit image open as well. I then applied exactly the same processes to both files.​
    Well you are applying the new edits in 16-bit despite the pasted 8-bit per color data and further, are you using adjustment layers and then flattening?
    I'm not sure what you're after. You've got a hight bit capture device. Why would you reduce the data provided to you?
     
  8. digitaldog

    digitaldog Andrew Rodney

    If you want a technique to use in Photoshop to visually see differences in edits you make to identical documents, try this:
    http://digitaldog.net/files/Apply_Image.pdf
     
  9. Logic dictates that raw 14-bit files saved out to 16bit should retain more fine tones compared with a "clipped" 8-bit.​
    The 14-bit part is in reference to your camera's Analog to Digital converter (ADC) ability to sort out of the electronic signal dump from each pixel sensor site the best and cleanest (with regard to noise levels near black) pixel detail using 14 bits selectivity which is now locked into the digital version of your file.
    The 16-bits once it's in the Raw converter is merely interpolating extra "sorting handles" (bits of precision) for the instructions provided by your editing tools to more precisely provide a smooth preview response when you edit your images on an 8 bit video display pipeline.
    You can't readily see these bits except through the precision in your edits which show up as smooth gradients that don't posterize as much compared to editing in 8 bit. Once you convert to 8 bit you've thrown out all those handles/points of precision for mapping further future edits that show up in the preview.
     
  10. digitaldog

    digitaldog Andrew Rodney

  11. Many thanks for the responses guys. Tim & Andrew, apologies if I've overcomplicated the question. What I'm trying to get at, is whether a supposed 14-bit raw file I'm opening up in Photoshop is, in fact 14-bit upped to 16, as I see no difference in results between the "upwardly interpolated 14-bit" and "upwardly interpolated 8-bit" files. (Hence my experiment: opening the original straight off as 16 bit, and opening an 8-bit version of the original, then interpolating it to 16bit). It's almost as if the original 14-bit data only really exists in the Nikon proprietary domain, (ie. View NX), but gets chopped as soon as you export it, regardless of whether you're saving it out to 8-bit, or a 16-Bit TIFF. How else would the filters behave identically on both?
     
  12. ...so if anyone wants to have a little investigative fun, try the following.
    1) Open up a NEF file in Nikon View NX. Preferably with light pastel shades across wide areas.
    2) Export the file as 8-bit TIFF from View NX.
    3) Then export the file as a 16-bit TIFF from View NX.
    4) Open the 16 bit TIFF in photoshop.
    5) Open the 8 bit TIFF in photoshop.
    6) Apply your choice of posterisation filters (comment 5, as per Andrew's suggestion) identically to both files. At this stage, there are obvious visible differences. So far so good. 16-bit looks better than 8-bit.
    7) Close files without saving.
    8) Open both again, the 8-bit and the 16-bit
    9) Now, interpolate the 8-bit upward, by cutting and pasting it into a 16-bit window of the same dimensions so you end up with two identically sized 16-bit versions of the same image, one being the original 16 bit, the other an interpolated 8-bit
    10) Apply the same choice of posterizing filters.
    Is there any difference between the two? Remember, one is meant to be the 14bit interpolated to 16 bit, the other only an 8-bit interpolated into 16 bit.
     
  13. Once long time a go I was making levels adjustments to a photo with lots of clear blue sky. First I tried levels adjustment with 8-bit photo and it resulted banding in blue sky. Then I tried levels adjustment with 16-bit photo and it resulted acceptable image.
     
  14. Ron - I'm interpreting well beyond my competence level here, but looking at your steps 8-10 (and comparing to the steps before), it seems to me that what is happening is normal. The banding or other artefacts that you are getting are occurring is because this particular dataset/workflow is sensitive to errors introduced during the posterizing. Put more simply, the original data is less of an issue than the 'quirks' of the mathematical processes (like rounding) when using an 8-bit process. (I'm hedging by referring to dataset/workflow because it may be different with different image and filter combinations).
    I _think_ if you want to test my hypothesis, you could confirm this way: in your step 5) above, (up)convert the 8-bit tiff into a 16-bit tiff after opening. Then do the rest. I'm guessing you will no longer see the differences that you saw in step 6, because now the same filters will have enough 'wiggle room' (pardon the technical terminology) to work without introducing the differences you saw.
    It would probably be possible to come up with a set of tests/filters that would show differences between multiple combinations of original 8- and 16-bit images worked on in 8- and 16-bit space. But sometimes differences just may not be visible.
    (Sorry I can't test the particular things above in the coming days...)
     
  15. digitaldog

    digitaldog Andrew Rodney

    Your capture is 14-bit. Photoshop considers all data that is greater than 8-bit per channel what it calls "16-bit". There is no reason for it to show you anything else since it's high bit data. It treats that data accordingly as high bit (actually 15+1 bit but that's another story).
     
  16. Andrew, you didn't catch Ron's experiment?
    This topic is not about understanding the technology in order to make beautiful images. It's about analyzing the behavior of high bit data viewed on an 8 bit video system and correlate anomalies in that low bit preview using a non-real world posterizing treatment to show inconsistencies in the significance of working with high bit data.
    Not only is it not a scientific analysis but it doesn't provide new information a photographer can use to make better photographs.
    News flash...the preview you see is not a real object subject to scientific scrutiny to determine repeatability across a wide range of captured photographic data. It's all an interpretation by software whose designers are not going to tell you how they arrive at their version of mapped digital sensor 1's & 0's on a display. The camera is not a scientific instrument nor is the software used to draw the data on the screen.
     
  17. digitaldog

    digitaldog Andrew Rodney

    It's about analyzing the behavior of high bit data viewed on an 8 bit video system and correlate anomalies in that low bit preview using a non-real world posterizing treatment to show inconsistencies in the significance of working with high bit data.​
    I see nowhere the OP said anything about using an 8 bit video system, I see no mention of the specifics of his hardware. I do see where I suggested the banding is quite possible the result of the video path and not the image data.
    Not only is it not a scientific analysis but it doesn't provide new information a photographer can use to make better photographs.​
    Again, says you. It appears just the opposite, no questions or comments about aesthetic attributes making a better photo, all about a scientific analysis relating to bit depth and banding. Everything Ron stated is about scientific analysis and has nothing I can see with subjectively to make better pictures. But based on our recent conversion on LuLa* where actual scientific colorimetric and non ambiguous analysis disproved your misconceptions about CMM's (ACE specifically) it's understandable you would suggest this post isn't about science but making pretty pictures.
    *http://forum.luminous-landscape.com/index.php?topic=98048.msg803502#msg803502
     
  18. What does the OP want, Andrew? Do you know? Have you conducted his experiment?
     
  19. digitaldog

    digitaldog Andrew Rodney

    What does the OP want, Andrew? Do you know? Have you conducted his experiment?​
    Whatever he wants Tim, I have to believe it's something based on sound scientific methodology, something you've recently gone on record ignoring and dismissing (hence the URL).
    Have you conducted his experiment?​
    Have you?
     
  20. I don't need to because I make images, not posterization tests. Which do you do?
    Or do you like dominating forums making sure you are seen as the authority on a subject and pouncing on someone who questions you on some minute piece of technology no one gives a sh*t about.
     
  21. digitaldog

    digitaldog Andrew Rodney

    I don't need to because I make images, not posterization tests. Which do you do?​
    Of course you didn't nor need to, interesting you asked if I did. Doesn't stop you from making statements about the processes. As to your question, I do both. Both ARE possible. Obviously that comment infers that the OP isn't making images and his questions are silly since it's about the 'science' of imaging of which you apparently have no interest or thought understanding.
    Or do you like dominating forums making sure you are seen as the authority on a subject and pouncing on someone who questions you on some minute piece of technology no one gives a sh*t about.​
    The dog isn't here or LuLa to establish dominance but rather point out the fallacies and misinformation some posters like yourself feel are factual while not providing a lick of evidence to support your claim, as you did with your posting about the Adobe CMM.
    Tim, you can continue to make what you feel are lovely (subjective) images. That keeps you free from having to get your facts straight. You are not immune or protected from making posts that have no basis of fact and expect that others will not call you out on the misinformation!
    BTW, thanks for speaking for everyone and incorrectly suggesting no one contributing thus far (yourself excluded) gives a sh*t about. If that were true (and it isn't like your comments about ACE), this discussion would never existed in the first place.
    Now back to the actual topic if that's OK with you?
     
  22. Greg, Andrew thanks for your comments - that is exactly my point. Both files are up-sampled to the same 16-bit photoshop space. If I paint a gradient at this point and start playing with it, the *painted gradient part* of the image is obviously smoother than the same process repeated within an 8-bit space. I can see the superior 16-bit result in line with my expectations and the difference is obvious.
    BUT... what I don't get is the actual captured photo itself.
    Forgetting filters for a moment, if I take an 8-bit photo and convert to 16-bit, next take a 14-bit photo and convert to 16-bit (which is effectively what I did in the steps above), would it be reasonable to expect the photo itself to hold more data/subtle tones "hidden" within the image? If I then applied the same extreme posterization side by side, to the point where both images are pushed to reveal banding and grain, should the converted 14-bit version not be at least slightly better than the one originating from an up-sampled 8-bit? In other words, if the camera originally "sees" and records 14 bits worth of tonal variation, should the resulting high-bit TIFF not be visibly superior to an up-sampled 8-bit, following all the editing?
    I cannot see any difference: it is as though both sets of outputs (the 8 and the 16 bit TIFF) contain identical amount of actual image data. The "16-bit" TIFF which I am outputting from Nikon simply appears to have 6 bits chopped off, then 8 bits of extemporaneous data appended to it. Hence, when both are edited side by side in 16-bit space, there is zero difference. This would seem to indicate that somehow the RAW data is internally clipped to 8-bits before it even "leaves" the Nikon domain, regardless of which TIFF bit rate you ultimately choose.
    As I said originally, maybe I'm missing something here. It's not gonna ruin my sleep but if I'm right, then folks have been exporting 8-bit images to photoshop all along, mistakenly assuming they are "high-bit" when in fact, the only "high-bit" elements are Photoshop-generated artefacts like gradients and effect filters - NOT the actual images you start off with.
    Please prove me wrong...
     
  23. digitaldog

    digitaldog Andrew Rodney

    Forgetting filters for a moment, if I take an 8-bit photo and convert to 16-bit, next take a 14-bit photo and convert to 16-bit (which is effectively what I did in the steps above), would it be reasonable to expect the photo itself to hold more data/subtle tones "hidden" within the image?​
    I believe it is reasonable to expect that the 14-bit original has and continues to supply more tones than the 8-bit version "up-sampled" to 16-bit. At this point, anything you do within the high bit files could be equal such as creating a gradient or of course, pasting additional high bit data into that document. But the 8-bit data, even with the additional padding of being in a high bit document doesn't create more tones out of thin air. I have heard in the past, some report that IF they are handling data that is 8-bits per color, they see less banding editing after converting to high bit and applying the edits. Within reason. This is what ACR and Lightroom do FWIW. Even with 8-bits of original data, all edits are applied high bit within the processing.
    If I then applied the same extreme posterization side by side, to the point where both images are pushed to reveal banding and grain, should the converted 14-bit version not be at least slightly better than the one originating from an up-sampled 8-bit?​
    Converted 14-bit? If you mean should this original high-bit file suffer less data loss than the 8-bit per color version, with equal edits upon the data, yes. That is exactly what you should see with a high bit file versus one that isn't to the point that the rounding errors produce issues with lower bit depth data.
    The "16-bit" TIFF which I am outputting from Nikon simply appears to have 6 bits chopped off, then 8 bits of extemporaneous data appended to it.​
    I suppose that's possible, I'm not a Nikon shooter so I can't fully follow Ron's steps. Don't have the files or the Nikon software. You might want to supply an NEF to test. It might be interesting to test this without the Nikon software, just ACR or Lightroom.
     
  24. ..erm btw thanks for your comments Tim, actually I am interested in beautiful images. Which is why I want to ensure I understand my equipment and software as thoroughly as possible. I am not using an 8-bit video display, and I'm not out to boast about my equipment.
    What I will say is that since I do not have the luxury of owning a darkroom or a large format film camera any more, I am trying my best to yield the best result in B&W photography using digital as I possibly can. These may well be "minute piece of technology" but actually, I do give a sh*t about it - if that's OK with you.
     
  25. Ron - I agree with you overall, but my point (in part) was that whether or not the difference is visible will depend on the original image and the filters applied. I think in your specific case it's that the 'stress' that leads to artefacting is more a function of the workspace (high-bit is what I meant, although I wrote 16-bit). In other cases, it may be the data limitations in the original image (for example, it is possible - if unlikely - that an image could have exactly the same colour/brightness values for a large area) althoug it is less likely for a high-bit image.
    That said, there may be other things at play here, like the specific workflow. Is there any reason not to try opening/processing the original raw files in photoshop?
     
  26. I am not using an 8-bit video display, and I'm not out to boast about my equipment.​
    I didn't think you were boasting about your equipment. I'm referring to an 8-bit video card pipeline within your OS's system. I'm assuming you're talking about your display's high bit internal calibration LUT hardware which still only receives 8-bit video through your video card's frame buffer unless you're using a video card that supports higher bit throughput.
    From reading your original post and title indicating the word "Help" I was trying to determine whether you were attempting to sort out a video driver bug that you suspected was not showing through previews what you thought should be high bit data behavior according to how you understood high bit editing workflow with normal editing. Or whether you were trying to understand how high bit data could be detected in a preview through the use of a posterization test which you asked us to conduct on our own system. I either didn't see the point of it or understand what you were checking for with regard to seeing the effects of high bit data on a low bit video pipeline which made me wonder just what you needed "Help" on.
    These may well be "minute piece of technology" but actually, I do give a sh*t about it - if that's OK with you.​
    That statement I made was directed at Andrew's harping in a LuLa discussion he and I had which he provided in his link on what I referred to as "a piece of minute technology" such is the ACE CMM that operates under the color management hood hardly any photographer really pays much attention to or "give's a sh*t about".
    Andrew has a long history in online discussions of overcompensating in order to get his point across by linking to other discussions in order to make the other person he's arguing with come across as someone whose information can't be trusted. I tend to ignore him at that stage because I hate seeing a grown man who already has a long history of establishing himself as a trusted expert lose it by going to those lengths over something so trivial.
    I still don't know if you got the help you requested.
     
  27. OK, I think we're getting somewhere.
    I just tried opening the NEF directly from photoshop. It seems that after heavy posterisation, a very tiny difference occurs between the original 14-to-16-bit and the up sampled 8-to-16 bit. Too tiny a difference to say "wow, this is clearly superior", but a difference nonetheless - confirmed in the histogram.
    View NX continues to produce identical results. This would suggest my earlier notion that View NX cuts the data before exporting it...
     
  28. Ron, thanks, that's interesting. I admit that I have rarely touched ViewNX - so while I don't want to comment on whether that's something in the software or some setting, it does identify the point where something odd is happening.
     
  29. digitaldog

    digitaldog Andrew Rodney

    That statement I made was directed at Andrew's harping in a LuLa discussion he and I had which he provided in his link on what I referred to as "a piece of minute technology" such is the ACE CMM that operates under the color management hood hardly any photographer really pays much attention to or "give's a sh*t about".​
    Translation: Andrew (with agreement from others equally knowledgeable on the subject) have corrected the mistakes I, Tim Lookingbill have made. I'd rather not make this a learning experience and gain knowledge from Andrew and those who agree with him who corrected me so I'll ignore and not give a sh*t about my errors.
    The most recent example was just yesterday with respect to how CMM's work. More examples are available...When Tim's wrong and corrected, it is therefore (in his mind) minutiae that (Tim speaking for everyone), no one cares about.

    Getting back (again) to topic, I think Ron you may have found the problem here, the Nikon software. And yes, the differences shouldn't be huge unless you inflect significant edits upon the data till you reduce the levels (as I suggested using Posterization) to the degree you see the effect on-screen. Depending on the output device and image, color gamut, degree of edits etc, this could show up on a far more demanding output. With high bit data, you don't have to worry about this happening using sound image editing techniques; there's just so much extra data padding the process.
     
  30. It seems that after heavy posterisation, a very tiny difference occurs between the original 14-to-16-bit and the up sampled 8-to-16 bit. Too tiny a difference to say "wow, this is clearly superior", but a difference nonetheless - confirmed in the histogram.​
    How many discrete levels of posterized separated tone are you analyzing to arrive at that conclusion using this test?
    Heavy edits to high bit data is only useful when applying to high resolution pixel levels of tones spread across a 255xRGB preview which employs sophisticated math to map changes to these fine, discrete tones through edits on millions of color combinations so I'm having a hard time understanding what you're learning from this test.
     
  31. (to be continued)
     
  32. I have been giving this general issue quite a bit of thought lately, and I have also done some numerical experiments.
    Let me focus my discussion on the issue of banding, and to further focus the discussion let us consider two cases, one in which an image is acquired in 8-bit mode and the other in which the image is acquired in 16 bit mode. Let us futher consider only the case of visibly detectable banding occurring in a smooth gradient. Let us further assume for the sake of discussion (without proof) that as a practical matter no matter how much manipulation you perform on an image acquired in 16 bit mode there will be no visible banding.
    Whether you will see banding occurring in an 8 bit image depends on a number of factors. For example, if you computer-generate a perfect gradient and express the pixels amplitudes in 8 bit words, and we then subsequently perform heavy manipulation of the gradient there is a possibility to generate visible banding. However, this case is probably not very interesting to photographers because most of us are not very interested in images generated by a computer.
    OK then, let's consider images acquired photographically, and assume that the object being photographed is a smooth gradient. If there is no noise in the measurement chain then banding is a possibility, particularly if the image is heavily manipulated.
    However, something entirely different happens if there is noise in the image processing chain (e.g. film graininess and sensor noise in a film/scanner system, or sensor noise in a digital camera) then something entirely different happens. If the RMS noise is comparable to the step size in the digital to analog converter then the gradient will be smooth (i.e. no banding), but there will be some noise in the image. This property is preserved under heavy image manipulation, provided that the 8 bit image is converted to 16 bits prior to image manipulation. (I suppose things might be a little different if the image isn't converted to 16 bits prior to manipulation, but I won't consider this case.)
    It actually doesn't take much noise to assure an absence of banding. About 1/3 bit of RMS noise is enough to basically eliminate banding. For those who care much about signal processing, what we have been discussing is basically the concept of dithering, and it is a well-known method of improving certain aspects of performance at the cost of introducing a little noise into the result.
    One other thing. If the noise of the system is comparable to the step size of an 8 bit A/D converter then there is very little benefit in using a higher resolution A/D converter, and if the noise is much bigger then there is virtually no benefit.
     

Share This Page