The Nature of Noise

Discussion in 'Digital Darkroom' started by lobalobo, Apr 30, 2009.

  1. There has been a debate in DPReview (and some discussion here) about whether downsampling an image taken with a high pixel density camera can effectively eliminate the noise inherent in such images, restoring the benefits of low pixel density. Those who say that downsizing won't work as well as shooting with a low pixel-density camera in the first place argue that because the noise created by high-pixel density is not independently distributed through the image, downsampling won't work very well; those who disagree say that while this may be true for downsampling a jpeg image, if one shoot RAW downsampling works just fine. The argument here seems to focus on digital artifacts (grainy images, e.g.) but I have questions about the ability of downsampling to achieve other benefits of low pixel-density:
    - First, I have understood that because low pixel density permits larger photosites, dynamic range is greater in a low-pixel density camera, and that it is these larger photosites (not sensor size) that contributes to the advantage in dynamic range. If this is right, downsampling from a high-density, small photosite camera is not going to help on dynamic range, right?
    - Second, there seems to be superior tonal qualities associated with low pixel density cameras (noticed across the range from MF digital images that look like LF film to the better colors from older, lower mp point-and-shoots compared to the newer, higher mp version). My understanding is that this is a product of lower noise, not the kind that produces grain-like defects in an image, but the kind that detracts from the purity of a hue. Can this sort of noise be corrected by downsampling? I wouldn't think so, but I am curious.
    Thanks in advance.
     
  2. .
    Opening at 1/2 bypasses the need to demosaic and all the guessing that Bayer demosaicing programs inflict on an image. If that's "noise" then you've avoided it by opening without demosaicing. Free http://www.irfanview.com/ and others do this as an option.
    Note that I tend to upsize 2x (at least) before JPGing to try to bypass the 4-pixel compression grouping effects and detail loss due to artifacts, so much of my Flickr stream full-size images are actually larger pixel counts than the Raw, but smaller file storage byte counts than the Raw, even at 0 compression 100% quality -- okay, and JPG is also 8 bit per RGB pixel rather than 14 bit Raw.
    You're dealing with unresolvable preferences and speculation on what technical details influence the final image results, and I imagine you will never convince anyone to change their minds, but you will learn a lot about the inner workings of your photo gear, and that's a good thing.
    First of all, there's no such thing as simply "noise". Can you be more specific? Once anyone gets more specific, then research and hands-on testing can be done to understand (a) if that noise exists, (b) if it matters, and (c) what anyone can do about it.
    "... Eliminate the noise inherent in such images ... " is a preposterous presumption implying there is no equivalent or worse noise inherent somewhere in the other systems being compared. I've yet to see an "all OTHER things being equal" test since everyone changes more than one thing at a time, such as when comparing different size or pixel density sensors, they are also comparing different antialiasing filters, different sensor array chip and support chip specifications, different lenses, different camera electronics, different internal and external software, and so on. So, what's the point?
    System (not individual parts) testing to YOUR preferences is all that matters, not comparing gear to gear.
    And MTF is multiplicative, not additive or subtractive -- the system is the product of all components, not limited to one supposed weak link.
    .
     
  3. Opening at 1/2 bypasses the need to demosaic and all the guessing that Bayer demosaicing programs inflict on an image.
    This doesn't seem to make sense, unless you mean something completely different from what you seem to mean. I don't believe that there is a way to get from the RAW data to a meaningful image without "demosaicing" the Bayer-array produced data.
     
  4. Like Sam, apparently, I did not understand Peter's post. Not sure in what way Peter intended to respond to my original post, but thank him for trying ("preposterous" or otherwise). Anyway, the DPReview article to which I referred is here:
    http://blog.dpreview.com/editorial/2008/11/downsampling-to.html
    In response to this article, critics claimed that it applies only to JPEGs, not RAW files, which can be downsampled more successfully. And my questions were simply whether this supposed more successful downsampling could affect dynamic range or color tone.
     
  5. Like I said in another thread about raw and this includes jpegs as well once the sensor data has been written onto the memory card, it's all interpretation from there according to the software forming the image on screen and this includes the quality and/or quantity of noise.
     
  6. I looked for Tim's "another thread" - I don't know if this is the one he meant, but what he did in this one seems to be a lot more effective (by preserving original image size, if nothing else) than the averaging-pixels-by-downsampling approach:
    http://www.photo.net/digital-darkroom-forum/00SXz1
     
  7. .
    Sam and Lobalobo, thanks for your insights. Let's read the references to see what we can make of them.
    ----------
    Sam, if a demosaic program tries to create 4 NEW RGB values for each original R, G, G, and B pixels, for cells that originally contain only an R value, or only a G value, or only a B value, some of which may be inaccurately misrepresentative of the original image due to recoded or playback system noise, which may then be amplified from one errant pixel to influence the surrounding 8 (or more) pixels with system noise, not with original image information, then instead, simply opening that image at 1/2 dimensions automatically combines the 4 separate RGGB values into one RGB value without demosaicing algorithms, and is a quick way to see a Raw image and not worry about the intrinsic problems with whimsical antialiasing filters and demosaicing algorithms. Just read:
    ... to explore some of the possible sources of image degradation (noise?) we immediately bypass by not using ANY of these demosaicing algorithms at all.
    ----------
    Lobalobo, regarding the DPreview link:
    ... (thanks for referencing it now ), right up front, it clearly says,
    "... downsampling four pixels into one averages noise ... "​
    ... which is exactly what I wrote in my first reply, but somehow you both felt surprised by information from me that this is OLD NEWS and many of us have been using Raw converters that do this at the front end of Raw development for speed and clarity, especially when we don't need the full capture pixel dimensions for a particular purpose -- web sharing, email, proofing, and so on. What the article fails to mention is:
    "... demosaicing amplifies noise ... "​
    The difference is between averaging noise and amplifying noise, and so, downsampling averages , demosaicing amplifies , and so, yes, downsampling, by comparison, reduces noise, hence my suggestion to do both -- avoid demosaicing and downsample directly from Raw -- double win-win in noise reduction, at least by comparison.
    However, the article goes on to downsample and compare AFTER an unspecified demosaicing system has amplified and added "noise" now integrated into the non-Raw image files (and each Raw developer has it's own "noise" along with the results of it's own demosiacing algorithm).
    As an alternative, I drew your attention to downsampling BEFORE demosaicing, INSTEAD OF demosaicing, directly from Raw. Neat, eh? Try it and let us know if you like it for occasional purposes. I use it all the time. Free in:
    ... FOR A REASON -- because it's quick and clean, because it doesn't waste time demosaicing (which may add "noise" of it's own, and may misinterpret and may misamplify noise, and may distort original image RGB values), and as suggested, it averages existing noise without adding or amplifying noise.
    That was the topic, right? In spite of the thread subject line, the topic is really,
    • Can noise be reduced by averaging downsampling?
    ... and the answer is yes, of course, especially since noise reduction is already being implemented by some algorithms that average without downsampling, and even more so if done without demosaicing . If you're gonna end up with a smaller file anyway, why waste time demosaicing and inaccurately calculating the larger file information to begin with? Why not go directly to the 1/2 size file from the original Raw? Seems obvious, eh? And it is, and it has been a native option in Raw converts from a variety of sources since the beginning of modern Raw digital photography -- just news to some of us, and hey, news travels in cycles, and old news feels new in another season or two even if some of us never forgot!
    More importantly, Lobalobo, can you tell us more? What's important to you in your real-world photography here? What have you got now that's not pleasing? What do you want that you can't create now? Please, tell us more!
    .
     
  8. "Sam, if a demosaic program tries to create 4 NEW RGB values for each original R, G, G, and B pixels, for cells that originally contain only an R value, or only a G value, or only a B value, . . . ."
    I tend to think that there must be a demosaic process involved in any event, regardless of whether it is a translation of each RGB pixel or a translation of combined RGB pixels. Some interpretive algorithm must be used to accomplish the translation from the luminance value of individual pixels combined with the color filter assigned to each pixel to an actual RGB image, also composed of pixels. The fact that there may be fewer pixels in the resulting image than in the source data doesn't change this. It could even be argued that a more complex interpretive process is required to produce a smaller number of resulting pixels from a larger number, rather than a more-or-less one-to-one translation.
     
  9. .
    Earlier: "... It could even be argued that a more complex interpretive process is required to produce a smaller number of resulting pixels from a larger number, rather than a more-or-less one-to-one translation ... "​
    Don't waste you time arguing about a more complex interpretive process -- it's called averaging because it actually is averaging , a rather simple process, and it's averaging down -- 4:1.
    Alternatively, any and all of the dozens of varieties of demosaicing algorithms (read the Wikipedia and other resources) are waay more complex and are trying to expand one pixel to influence 8 or more neighboring pixel values -- 1:8 at least, the opposite of downsampling, at least 2 times more complex (1/4 x 8/1 = 2x) and as much as 32 times more complex (4:1 x 1:8 = 32x). Try it and time the difference yourself on your own computer.
    Anyway, if the user is happy with a smaller file, skipping demosaicing is a win-win choice. If the user wants a large image file that compares with the original, then:
    ... and other resources are probably best implemented.
    And, no, my Raw converter has all sorts of options, including original grayscale with no RGB values, and original RGGB map undemosaiced, both of which are fast since, of course, there no demosaicing . Demosaicing is only needed for full size RGB, nothing more. 1/2 size load is the fasted WITH full RGB values, however, and of course, no demosaicing.
    (Why don't we get on a conference phone?!?)
    .
     
  10. Thanks to all, but ... Peter, I was not surprised by anything you said, becasue I did not understand much of what you said (save your repeating that downsampling of RAW is better) even though I did understand what Sam has said in response to you, and do understand what was said in the DPReview link I provided, and of other criticisms of the DPReview link.
    Also, as I feared, this thread has become a rehash of the original issue, when what I was hoping for is an explanation of how downsampling might affect dynamic range and tone. Beggars can't be choosers, though, as they say, and I appreciate every response in any case.
     
  11. "Don't waste you time arguing about a more complex interpretive process -- it's called averaging because it actually is averaging , a rather simple process, and it's averaging down -- 4:1."
    Even then, some interpretation would still be required - the program must select which 4 pixels are to be averaged, and so on through the multi-megapixel array. It wouldn't be interpolation, but it would be interpretation.
     
  12. Peter-
    That being said, your point does lend some weight to the argument that a reduced-pixel image would have lower noise per pixel (on average - due to the averaging process) than the original data, and your point about the additional artifacts added by the de-mosaicing process makes sense also. I'm not convinced, however, that this approach would result in a higher-quality image for most purposes for which we produce images, as opposed to simply doing a quick, low-resolution peak at the image on a computer monitor. It seems to eliminate most, if not all, of the commonly-understood advantages of shooting in RAW format.
     
  13. .
    Also, my Raw converter has 4x (or more) upsizing directly from Raw, and that means each pixel value is used to influence at least 12 neighboring pixels -- 1:12 -- and probably more, perhaps 1:64 to 1:144 pixels, and I might as well go to lunch as it calculates that (I do it in batches and go away, sometimes overnight)!
    Demosaicing is fun, but not necessary. And the "interpretation" is called simple mathematical averaging , not complex at all.
    Sometimes the simplest answer is actually simple!
    And, if you want to understand what I wrote, just try it yourself -- hands-on personal experience is everything:
    Now, if DPreview had tested Sigma Foveon RGB images that need no demosaicing in the first place ... ;-)
    And that's the point -- downsampling by the example at DPreview was complicated by their lack of discipline and understanding of the complications induced by various demosaicing problems, which I immediately suggesting bypassing. That is all.
    .
     
  14. Lobalobo-
    Here's a thread that really goes into the effects of downsampling on dynamic range - I don't know how much, if any, of this discussion is valid, but it certainly sounds impressive:
    http://luminous-landscape.com/forum/index.php?showtopic=30658
     
  15. .
    Sam, I think we often get overwhelmed comparing things to things , rather than comparing things to our needs .
    Capturing in Raw permits the greatest potential for subsequent reinterpretation and representation, that is all (and has nothing to do with this thread -- this thread is NOT a dissertation on whether to capture Raw or not).
    If we also have a separate contemporaneous goal of an image that only needs 1/2 the size of the Raw original, then as luck would have it, we can get to that goal rather quickly during Raw conversion by simple 1/2 averaging without demosaicing.
    Otherwise, we can open the Raw and demosaic it at any size (I often work at 2x), and have our way with the image to our heart's content, and save at 16 bit PSD or TIF for later consumption. Since that's a "special" interpretation, we'd shrink a COPY of that to our reduced target, rather than reopen the original Raw (since the original Raw does not know of our recent tweak preferences).
    Also, in these threads, we can get lost arguing minutia that has no effective import or impact on real-world effectiveness. Shrinking an image to 1/2 size before or after demosaicing is probably immeasurably unimportant for most people who need only 1/2 size copies, so what does it matter if one way is quicker and produces cleaner images than the other? Quicker and "cleaner" by what degree? DPreview is unable to say since they had no idea what they were doing -- they had no real world target. Too bad.
    Heck, a slight change in JPG % compression has more impact than this whole topic on noise reduction via downsampling, so we're arguing angels on the head of a pin, not actually using the pin for effective sewing of a wearable garment, so to speak.
    However, I hope I've expanded everyone's awareness of alternatives they might not have considered, and if we need a quick batch of 1/2 size images for some distribution reason, there's a really quick batchable way to get there (IrfanView).
    And, again, Lobalobo, what's your real world challenge? Improved image qualities at original size, or, various ways to present a smaller size image, and what happens to (and how to control) various image qualities on the way there, or, other? Please, tell us more about what you are trying to accomplish, and what's become an insurmountable struggle for you.
    .
     
  16. Interesting, but in a sense the entire discussion is moot. Modern (is there any other kind) DSLRs have excellent noise characteristics in all real-world photography - and the noise that is present is not only minimal (and often essentially invisible to viewers) but less than the film equivalent expressed as grain.
    Dan
     
  17. @Peter
    Opening at 1/2 bypasses the need to demosaic and all the guessing that Bayer demosaicing programs inflict on an image.​
    Forgive me if I have misinterpreted what you have written, but you seem to be saying that one can obtain a color image from a raw file without employing a demosaicing algorithm.
    I can easily understand that opening a grayscale raw image at 1/2 will average the values and thereby, quite possibly, average (reduce) noise. Having done that however, you still end up with a grayscale image with each pixel represented by a single (average) value. How do you get from there to a color image where each pixel must contain a triplet of values (R, G, & B) without demosaicing at some stage?
     
  18. Thanks, Sam, for mentioning the dynamic range question in the original post. I'll look at your link. In the meantime, what about color? I was told by a photographer (and regular contributer to Luminsous Landscape) that the reason MF digital backs have a richer tonal range than smaller sensors is that the lower pixel density on these backs permit color to be recorded more purely, with less noise that could change the hue. If this is right--and it may not be, as I could have misunderstood--the question arises whether downsampling from a high-pixel-density camera can fix this problem.
     
  19. There has been a debate in DPReview (and some discussion here) about whether downsampling an image taken with a high pixel density camera can effectively eliminate the noise inherent in such images​
    Assuming you're OK downsampling, of course it will help reduce noise as pointed out above due to averaging of pixel data. Its not even close to a cure, but it should help a tad. I'm surprised there would be a debate but considering the vast wasteland DP review forums have become, I'm not surprised.
    If you want to reduce noise and retain all the pixels, at the very least, expose properly (ETTR) if possible. The problem is, we generally get noise at higher ISO amplifications because we need more light to expose the image and ETTR kind of flies out the window. Give the option of a capture with noise, or no capture that's cleaner in terms of noise, its not a difficult decision.
     
  20. .
    • Average 4 separate R, G, G, and B values down (downsampling) to one RGB value is not demosaicing.
    • Reading the values regardless of RGGB position and reporting only the value without the RGGB map displays as grayscale, and is not demosaicing.
    • Reading the RGGB map and displaying the RGGB values as RGGB values is also not demosiacing.
    • Only guessing at the missing GB, RB, RB, and RG values to include in the known RGGB values is demosaicing.
    The easiest solution it to use Sigma Foveon direct RGB. ;-)
    ----------
    People think MF Medium Format cameras have nicer color (than what?). They need to justify their expense as if it were an investment. That is all. If I spent a lot for something, I'd say it was great(er), too.
    ----------
    When I say modern DSLRs, I'm thinking of post-2000. So I think of 1999-and-prior as perhaps exceptions to many rules, such as some DSLRs back then having multiple sensors and non-antialias (is that even a word?) RGB filters, storage on tape, less than 1 MP, and so on. That's all. Someone might do an historical study to find when things we take for granted first started, and when things we never heard of ended.
    ----------
    Lobalobo, I still am not sure I can teach myself what is your goal, what it is you want that you do not have, or, what it is that you have that you do not want. Is there more about your situation?
    .
     
  21. http://www.dxomark.com/index.php/eng/Insights/More-pixels-offsets-noise!
    http://www.dxomark.com/index.php/eng/Insights/SNR-evolution-over-time
     
    • Average 4 separate R, G, G, and B values down (downsampling) to one RGB value is not demosaicing.
    • Reading the values regardless of RGGB position and reporting only the value without the RGGB map displays as grayscale, and is not demosaicing.
    • Reading the RGGB map and displaying the RGGB values as RGGB values is also not demosiacing.
    • Only guessing at the missing GB, RB, RB, and RG values to include in the known RGGB values is demosaicing.
    Can't disagree with anything you wrote. You can average as many individual pixel values as you like; you still end up with only one (average) value per pixel. To represent a pixel in color it must have 3 values (one each for R, G, & B). To achieve that you must use some sort of demosiacing process, whether you call it guessing or calculating.
     
  22. .
    Earlier: "... To represent a pixel in color it must have 3 values (one each for R, G, & B). To achieve that you must use some sort of demosiacing process, whether you call it guessing or calculating ... "​
    Simple averaging and downsampling 4 separate R, G, G, and B pixel neighbors down to one resultant RGB pixel does not require demosaicing, and produces a 1/2 size image, and is very quick because it is simple averaging. That was the whole point of the thread from answer one on. How quick? As fast as changing a screen display image to 1/2 size on screen in any image viewing software -- simple averaging down. It wastes no time demosaicing.
    ----------
    Thanks for your insights, Ralph. Reading what's on the marketing minds of DxO is responsive to the opening post subject line. And, it helps pick apart the erroneous presumptions in the opening sentence:
    Earlier: "... There has been a debate ... about whether downsampling an image taken with a high pixel density camera can effectively eliminate the noise inherent in such images , restoring the benefits of low pixel density ... "​
    We've not really addresses the fallacy of thinking that high pixel density sensors must be noisier than low pixel density sensors. DxO's experience helps remind us that technology marches on, and the age old argument of how many transistors can we fit on the head of a pin never really held one answer for long.
    See also Zeiss's analysis of high pixel density in their Camera Lens News 30 and 31:
    • How to read MTF curves Part I:
    • How to see MTF curves in images Part II:
    .
     
  23. Peter says the following (among much else): "People think MF Medium Format cameras have nicer color (than what?). They need to justify their expense as if it were an investment. That is all. If I spent a lot for something, I'd say it was great(er), too."
    It is possible that you have an ulterior motive for everything you say, but it is a mistake that everyone else does. In my case, I own a $275 bridge camera, a couple of sub $200 pocket cameras, a $200 1950s press camera with a $175 125mm lens attached. In other words, I have not spent a lot of money on equipment; nor am I an expert. I have noticed though, online, in magazines that identify equipment, and in exhibitions I've been to, that almost without exception the digital images and large prints that I've seen that compare in dynamic range and subtlety of color with images that start on 4x5 film are taken with cameras that have relatively low pixel density (a possible exception being the medium, not low, pixel density, high-res MF digital backs, which produce wonderful dynamic range and colors too). This is what I've seen with my own eyes, not because I need to defend a purchase. I mention that I have had my impressions confirmed by others who know more than I, only because it is possible that any one observer can have an idiosyncratic perception. So accusing those with whom you disagree of dishonesty may be fun for you, but it's not always accurate.
    Peter also says (among much else): "Lobalobo, I still am not sure I can teach myself what is your goal, what it is you want that you do not have, or, what it is that you have that you do not want. Is there more about your situation?"
    Believe it or not, and apparently this is hard for you to believe, I am simply curious. I have no unmet goal. I was and am still wondering whether downsampling, which reduces resolution, can increase dynamic range or improve color. I've seen the debate over downsampling and noise and artefacts, and didn't intend to revisit that, but I wondered about dynamic range and color, which also strike me as advantages to lower pixel density, all esle equal. Some, Sam, e.g., have attempted to answer, and I am grateful. Others have jumped back into the original debate on downsampling and noise without addressing my specific questions about dynamic range and color, and that's fine, as of course I don't own the thread even as the original poster. You, though, Peter, have uniquely treated the post as an affront and have responded with disdain and insults (using words such as "preposterous" and accusing others of dishonesty in their opinions). Next time you are angered by a post, please just don't respond--would save yourself a bunch of time, too.
     
  24. .
    Earlier:
    • "... images that start on 4x5 film are taken with cameras that have relatively low pixel density ... "
    • "... online, in magazines ... in exhibitions ... digital images and large prints ... dynamic range and subtlety of color ... wonderful dynamic range and colors too ... I've seen with my own eyes ... "
    • "... dynamic range and color [from different sensor pixel densities] all else equal ..."
    • "... I am simply curious ... "
    Thanks for your insight, Lobalobo.
    I've seen this dialogic process many times before -- it take three or more iterations for the original poster to get to what's really on their mind. Not "The Nature of Noise, " but more like, "Where does color gamut, dynamic range, and signal-to-noise ratio in a presentation system come from? " And now I can address that (though any "4x5 film with low pixel density" I'll just write off to a typo, and "all else equal" when changing "only" sensors is an impossibility, so I'll ignore that, too).
    What you are looking at is the color gamut, dynamic range, and signal-to-noise ratio of the presentation medium -- screen or print -- and as such no one has any way of assessing the color gamut, dynamic range, and signal-to-noise ratio of the medium used to capture the original image (let alone the lighting, color gamut, and dynamic range of the original subject scene), nor can anyone use the final presentation form itself to assess the intermediate system used in between to modify and deliver that original capture to the final presentation medium.
    In other words, you're asking, "I see pictures that pop, is it the camera? ", and the age-old answer is, "No. "
    Any camera (and I mean ANY camera) can present a capture image to an image editing system that then manipulates and tweaks the image color gamut, dynamic range, and signal-to-noise ratio in ways untraceable to the original camera system, and then present that intermediate image for printing or display, which, of course, is then limited in it's own output color gamut, dynamic range, and signal-to-noise ratio of the display medium, but also by the display environment lighting, and by the eye and brain of the beholder (where color accuracy is not assessable, and off-white become white after a moment of viewing anyway, for instance).
    As for downsampling adding anything back in from the original scene that was missing, I think not. As for downsampling adding or reducing anything, probably only by averaging or by more complex algorithms different from each programmer, hence competition out there for our dollars, but whatever, we play with it for a pleasing resulting color gamut, dynamic range, and signal-to-noise ratio, and has nothing to do with original capture accuracy. Yes, workflow to get to that target is easier if each component is most accurate, or more punchy, or more of whatever the artist likes, but after the image is handed off at each stage, not much forensics can accurately identify whether what we are seeing compares more or less accurately to the original scene, if that matters. Heck, even Ansel Adams tweaked his output images to make his own desired impression where the system itself laid out a flat and boring image otherwise without his manipulations!
    Where does pixel density fit in here, other than as a miscible range of performance, changing with each sensor generation of new technology, season to season? Same with film, I guess, but who buys old film anymore when there's new film? Who buys cameras with old sensors anymore when there are new cameras with sensors? And what does it matter when we can fix it in Photoshop after all?
    Thanks for clarifying, and I've learned a great deal researching the side-track tangentials getting here. It's been a very provocative and productive thread. Thank you.
    .
     
  25. .
    PS - Lobalobo, what do you think of the DxO and Zeiss information resources presented in the links we shared? Do they help expand your understanding of what you might be seeing on presentation screens and in prints?
    .
     
  26. Thanks, the Dx0 sites are helpful on the general question, but not on my specific inquiries about dyanmic range and color. I haven't looked at the Zeiss link yet, but thanks for that.
    Regarding the substnace of your most recent posts, for which I am sincerely grateful, I do take your point that a digital imaging system can produce any outcome within its range regardless of the input. But this proves too much. In theory, I can take a number two pencil and write down a series of "0s" and "1s" that, when processed by a computer, and output device will closely resemble the Mozart symphony I'm listening to or sunset I'm looking at (or either, modified as I wish). That's in theory, but neither I nor any other human with a pencil knows how to write down the "0s" and "1s" in the right sequence, so we use microphones and cameras (and other equipment and related software) instead. What this debate is about, I take it, is how much better the microphones or, here, cameras are in knowing what to feed the digital imaging system.
    Turning to that question, I'm reminded that Richard Feynman once said that even quantum mechanics can be explained in simple, intuitive terms. I'm reminded of that quote when I teach (law) and I'm convinced that what is true of quantum mechanics is true of digital photography as well. So here are the two, simple, intuitive explanations I've heard for why large pixels produce better dynamic range and better color differentiation, followed by my questions:
    - On dynamic range: large pixels don't fill up with light (i.e., "clip") as fast as small ones, thus with larger pixels it's possible to expose for the shadows (i.e., let in a lot of light) without clipping the highlights. My question is whether downsampling from a high number of small pixels can replicate this effect. I'm guessing the answer is "no" because a clipped highlight (or black shadow) has no information to reveal upon downsampling. And I didn't think this had anything to do with noise. But I wasn't sure on either count.
    - On color, large pixels have less noise and so a standard "red" for exampe is going to be recorded by the camera closer to standard red when the pixels are larger. (And what I mean by "recorded by the camera" I mean record data that instructs the imaging device what color to produce if that color is withn the gamut of the imaging device.) With smaller pixels, the noise will either move the color from red towards some other color altogether, say green, or effectively mix in gray in the wrong proportion. My question is whether downsampling can fix the problem. I think not, because once the noise confuses the sensor as to what it has seen, so to speak, that information can't be recovered. But again, I'm not sure. (And yes, I know that the final imaging device can be told to produce whatever color the user wants, but the prospect of doing this manually brings me back to the listener or viewer with a number two pencil writing down "0s" and "1s".)
     
  27. On dynamic range: large pixels don't fill up with light (i.e., "clip") as fast as small ones, thus with larger pixels it's possible to expose for the shadows (i.e., let in a lot of light) without clipping the highlights.​
    Well that's simple and intuitive. However it occurs to me that an extension of this idea would lead one to believe that the bigger the pixel, the larger the dynamic range. I don't think that experience bears that out. Also, it seems to me that both large and small pixels fill up with light at the same rate (when you espress rate as a function of their total capacity).
    However it does make sense that a large pixel, filled to some small percent (x) of its capacity would give you a cleaner signal than a small pixel filled to that same percent. Now we're back to talking about noise, and I don't think you can discuss dynamic range without taking noise into consideration.
     
  28. Noise is the uncertainty in the data. Put another way, noise is the error in a given data value. The noise for a given data value is the difference between the recorded value and the true, but unknown value. Artifacts are not noise because artifacts are defined as non-random errors.
    Noise is inherent to the data. It can not be reduced or eliminated (but the misnomer noise reduction does sell software).
    Noise can be averaged (filtered). Data values with less uncertainty are averaged with data values with more uncertainty. The information content of the low noise data values is (must be) lowered in order to reduce the uncertainty in the high noise data values. Efficient algorithms used with skill often produce a useful compromise that improves the presentation of the data. But the total noise (uncertainty in the data) is not reduced. The total uncertainty in the image remains the same. If you can figure out how to increase the total information content of a data set after it has been recorded, you will become richer than Bill Gates.
    Understanding the effects of noise in the data during digital image processing is complicated as several types of noise are present. The Bayer de-demosiacing algorithms and other processes also average the data. Compromises must be made. However the fundamentals of information theory still apply. The noise content of an image can not be reduced or eliminated.
     
  29. - On dynamic range: large pixels don't fill up with light (i.e., "clip") as fast as small ones, thus with larger pixels it's possible to expose for the shadows (i.e., let in a lot of light) without clipping the highlights. My question is whether downsampling from a high number of small pixels can replicate this effect. I'm guessing the answer is "no" because a clipped highlight (or black shadow) has no information to reveal upon downsampling. And I didn't think this had anything to do with noise. But I wasn't sure on either count.​
    I've wondered about the reasoning behind this as well. I have heard it said that the larger the photosites, the larger the dynamic range. But as Mike points out, this isn't the be all and end all of the story. I also wonder whether the photosites ARE actually bigger. The fact that the 5D has a lower pixel density than the 1DsIV, does this mean that the photosites can actually handle MORE light or are they just a bigger in area?
    On your question of whether downsampling could improve dynamic range, I would have to think it would be a definite NO. As you speculated yourself, if the pixel is clipped in the first instance, no amount of down sampling is going to change that.
     
  30. Bernie-
    Here's a discussion of some of the issues you raise concerning sensor pixel size:
    http://www.clarkvision.com/imagedetail/does.pixel.size.matter/
     
  31. Mike Blume said:
    "Well that's simple and intuitive. However it occurs to me that an extension of this idea would lead one to believe that the bigger the pixel, the larger the dynamic range. I don't think that experience bears that out."
    There are other factors, but all else equal I believe that it is true, the larger the pixels the better the dynamic range.
    "Also, it seems to me that both large and small pixels fill up with light at the same rate (when you espress rate as a function of their total capacity)."
    Yes, but unless the camera's software knows to combine them into a single virtual pixel (which, I take it is what the new Fuji EXR technology does), they will all clip, I think.
     
  32. I make noise reduction software and I think I will chime in here.
    1- Better dynamic range and lower noise go hand in hand. Between clipping and black, it is noise that determines dynamic range. I would consider them the same thing. (As long as you're dealing with linear sensors, i.e. digital cameras, and not film.)
    2- Downsampling does reduce noise (and therefore improve dynamic range). Averaging values together will reduce noise and the averaging reduces it.
    This is also how image stacking works.
    3- Noise reduction plugins really do reduce a small amount of noise. How is this possible? It's because our physical world has certain properties (due to the physics of light and the nature of most objects) that statistically skews what the image should be. The ideal values for an image is not randomly distributed. Most/all noise reduction plugins will incidentally create values that are slightly closer (on average) to what the image should be, were it free of noise.
    Try this:
    Take images shot at two different ISOs, one very low (A) and one very high (B). Image C will be B with noise reduction applied.
    Do a difference composite to compare A-B and A-C. You should see that A-C is less different than A-B; it is closer to what an ideal image would be.
    4- The practical answer:
    You can ignore a lot of the math and statistical mumbo jumbo because it doesn't really matter. Humans don't see the world in terms of signal to noise ratios or standard deviations. There is a lot of sophisticated image processing in our brains that determines whether something looks 'right' or doesn't. We have models on how good something "ought" to be, but often these models aren't very good at predicting things and are therefore not that useful. At the end of the day, one of the most useful approaches is to simply look at an image and see if it looks right.
    a- The noise reduction plugins are the way to go (in my biased opinion). They (pretty much) can always help.
    b- You can't polish a turd! If an image is too noisy, the NR plugins cannot rescue them.
     
  33. Glenn says: "Between clipping and black, it is noise that determines dynamic range."
    I get the overall point, but definitionally, isn't dynamic range, loosely put, the distance between clipping and black?
     
  34. No, dynamic range is being able to distinguish separation of tone that make up shadow detail and highlight detail given the limits of a 8 bit 255 gray level video system and even more limited in a print.
    You have so many separate density levels out of these 255 levels for each RGB channel devoted to making up tonal differences in these extreme areas. When you start seeing flat foggy looking shadow regions combined with the stippled texture of noise, dynamic range width size and perception of detail starts to take a hit.
     
  35. I said: "but definitionally, isn't dynamic range, loosely put, the distance between clipping and black?" Tim responded: "No, dynamic range is being able to distinguish separation of tone that make up shadow detail and highlight detail given the limits of a 8 bit 255 gray level video system and even more limited in a print."
    Ok, but consider the following thought experiment: A wall contains, say, 24 equally wide vertical strips placed edge-to-edge within the field of view of two cameras. Strip 1 is Black and Strip 24 is White, and strips 2 to 23 are shades of gray that are progressively lighter moving from the first strip to the last. Imagine that the sensor in Camera 1 is exposed for the middle strips and produces an image that shows Strips 1 -5 pure Black and Strips 20-24 pure White, while the sensor in Camera 2 exposed in the same way produces an image that shows only Strips 1-2 as pure Black and only strips 23-24 as pure White. Are you saying that the sensor in Camera 2 does not capture a wider dynamic range than the sensor in Camera 1?
     
  36. No, dynamic range is being able to distinguish separation of tone that make up shadow detail and highlight detail given the limits of a 8 bit 255 gray level video system and even more limited in a print.
    I think most people would define dynamic range differently?
    Also, I personally wouldn't worry about quantization error caused from 8-bit coding if the image has noise in it, since the noise will naturally act as dithering and you won't see any problems with quantization error or banding.
     

Share This Page