How many usable megapixels will go on a full-frame sensor?

[craigd]

<p>Well, the comment I was replying to (by Dan South) speculated about a getting away from the 3:2 aspect ratio by a minor increase in full-frame sensor size from 36x24 mm to something like 36x30 mm. So my comment was, okay, but existing 35mm lenses may not work very well with that due to increased diagonal. (At the very least, I think you'd be likely to have strong vignetting in the corners with many lenses.) You could make larger lenses, but then you're alienating existing 35mm customers. So from a marketing perspective, how do you sell a minor increase in size like that? It's not an easy upgrade. It isn't all that much bigger, so there's no great increase in image quality. Your answer of "Uh yeah... it's called medium format" doesn't seem to me like an appropriate answer to me because MF, and even "tweener" formats like the Leica S2, are a great deal larger and a great deal more expensive than 35mm, and lenses designed for 45x30 mm or 645 are overkill if the goal is a minor increase like 36x30mm.</p>

<p>I can understand not wanting to take the time to read the whole thread if you weren't in on it from the start, but it's a weak excuse for mis-aimed sarcasm, a strategy which I see you are still trying with your snark about "I was actually out taking photographs" -- as opposed, I guess, to those of us who presumably have devoted our full time to this thread for the last two days. I really can't imagine how sad it must be to be the sort of person who thinks following this thread would require more than a few minutes now and then.</p>

[philip_wilson]

<p>Keith - if I understand your assertion correctly you are saying that there is no noise penalty to smaller pixel sizes. While your statement that both sensors capture the same noise is true the signal to noise ratio is a function of pixel size. If I remember my physics correctly (an I did once get two degrees in this topic) I believe the following is true<br>

The Signal to noise ratio of a photosite should improve with it's size (assuming the same technology and that the sensor S/N is phonon noise limited)<br>

The Dynamic range of the sensor is also limited, I think by the ratio of the signal to the sensor noise floor. Since the noise floor of most sensors will be similar while the signal in larger pixel size sensors will be stronger - larger pixels give a better dynamic range.<br>

Finally differaction effects will increase as pixel size falls reducing MTF (contrast)<br>

Since what we are interested in is the ability to resolve the wanted signal from the noise I would suggest that larger sensors will provide a better ability to do this. Indeed the weight of emprical evidence is with me on this statement. Comparissons of different technologies or generations of technology are clearly invalid. The one question I have is how far away are we from the theoretical limits and therefore what improvements remain. My understanding is that Backlit sensors are a significant improvement as the electronics obscure a significant proportion of the light coming into the sensor. I assume there are also other improvements in sensor and signal processing technology that will further increase performance.<br>

Daniel - while I do not dispute your claim that a sensor that out resolves the lens is worth having my point was simply that we are in the realm of diminishing returns and thus the real benefits of increasing resolution will not be realized. I did not state that the 7D was the maximum - just that I cannot see much benefit going further with current optics. I suggested 40 -60 MP on full frame was perhaps a reasonable limit but then went on to point out that at G10 and G11 densities this would be over 300 or 200 MP.</p>

[Uhooru]

<p>You're right Craig, 35mm lenses will probably not work with them, and who really wants to throw away a bunch of expensive lenses for a theoretical gain in IQ. I still think this whole thing is uber geeky and it seems this thread is longer than about anything I've ever seen here about photography itself, and I wonder why that is. But, again you are right Craig it was a snark, and I apologize.</p>

[Uhooru]

<p>Craig, one other thing, you first post said ". . . significantly larger. . ." that to me is MF. </p>

[keith_anderson7]

The theoretical limit, minus all the emotional responses that don't address your question, is that a ff sensor can hold about 8x10^14 transistors (800 trillion), so there is a long way to go. That number is based on the current geometries of microfabrication times the length and width of the ff sensor. If it's a bayer CMOS sensor (and it's likely that type will be eclipsed long before we actually hit the density), would be 1/4 that amount when you include the color channel filters (2 green pixels, one red, and one blue for every useful point). Since photosensors are generally bigger than transistors, it's going to have a scale factor for that as well, which may get you into single digit trillions.

 

But it's also likely there will be a paradigm shift (or two or three) long before we get into the Terapixel range. At least I hope. And in the meantime, the smallest geometries possible will also keep improving.

 

Maybe it's better to go out and enjoy the beauty of the world and of the people of the world (with your camera of course) and not worry about it.

 

Lenses will likely hit their limits before sensors anyway.

[sarah_fox]

<p>@Keith Reeder: What can I say, but "WOW!" (I'm being very restrained and polite with you, Keith.) The parenthetical phrase with which you take such violent exception is not particularly important to my point. Please re-read my post, and substitute any cost of higher pixel count that you wish, or simply stop reading before the aforementioned parenthetical phrase, and then perhaps you will understand what I wrote. I'm sorry, but I do not agree with you that higher pixel density results in reduced noise. However, per the original post, the discussion is one of resolution, not noise. I believe that is the factor I was addressing.</p>

<p>@Dan South: The automotive analogy is an interesting one, at least for me. I recently sold a very nice little sports car with all the bells and whistles. The dynamic stability control had a sensor go out, to the tune of $700. The side curtain airbag system might or might not have had a malfunction that, if it did, I cannot discuss. It turns out that the NHTSA files are replete with reports of spontaneous side curtain airbag deployments because of the ridiculously short analysis time required in a side impact and consequently the rate of false positives. These sophisticated safety systems are of questionable benefit to a older, sensible, country driver like myself. Most of my mileage is logged on 30 and 40 mph roads! Disgusted with that car, I sold it and bought an earlier technology (1970's) sports car without dynamic stability control and without airbags. Even though it's an antique, it's a simpler car that requires fewer repairs, and when repairs are required, the parts are simpler and cheaper. Best of all, I can drive my antique without the fear of being spontaneously attacked by a malfunctioning safety system I hardly need. My only point is that practically everything comes at a cost. Certainly higher MP images come at a cost, even if it's hard drive space, processing time, and the cost of upgrading to the next latest/greatest camera. If it's not something that conveys a benefit, it's not worth any additional cost, whether in photography or automobiles.</p>

<p> </p>

[sarah_fox]

<p>@Lannie: I get the intent of your question. I think there could be some surprises, though. The old "more MP is better" paradigm is growing old, and the down-resolution of Canon's G11 (re their G10) may be an indication of changes to come. If resolution were everything, then just consider the implications back in the film days. Wouldn't we all be carrying around 8x10 plate cameras? I certainly understand why a large format photographer wants such huge negatives, but there was a reason that I wanted such small negatives, back in my 35mm film days. You can't do the sort of photography I like to do with an 8x10 plate camera. That's not my thing. There will always be people going to extremes for the highest possible resolution, but that's not the mainstream of the market, as reflected by the relative numbers of 8x10 cameras vs. 35mm cameras.</p>

<p>In digital terms, I don't need a 50 MP FF camera body, and I don't think I ever will. My practices are inconsistent with that sort of resolution. So Canon can develop such a camera if they want (power to them!), and they can probably sell them to MP pushers. However, I don't think they will have the sort of mainstream success that will justify continuing to push in that direction. Instead, I want (and feel many others want) better low light performance and higher dynamic range. There's a LOT of room for improvement in both. Give me deeper wells, higher photon capture ratios, and better amps and ADCs, and I'll be there with my money. Nikon seems to have figured that out, and I think/hope Canon is not far behind.</p>

[zml]

<p>Sarah Fox><em> Give me deeper wells, higher photon capture ratios, and better amps and ADCs, and I'll be there with my money. Nikon seems to have figured that out, and I think/hope Canon is not far behind.</em></p>

<p>Hmm... Nikon (at best) has 1/2 to 3/4 f/stop of "less noise at higher ISO" on Canon due mostly to excessive, in my view, in-camera processing. Unlike most people making the claim you have made, I've shot in real life with all high-end Nikon and Canon digital gear (and most current MF digital backs) and this is my experience. The DxO Mark people don't agree with me but I don't agree with their conclusions either.</p>

[DickArnold]

I will not enter into this discussion except to support Sarah in her call for scientific rigor. I say this having been, in an earlier life, the dep. director of the R&D Service at the FAA and also the Product Team Leader for GPS and Navigation R&D and Acquistion. We conducted a lot of applied research. I see very little supportable research or real operational test and evaluation to support many of the assertions in this thread. I know nothing about sensor function, mechanics or electronics so I have no opinions on noise or resolution except what I anecdotally view coming out of my rather expensive and new printer. If one were to make judgments I would think they would come from controlled studies where known facts were indentified, variables were catalogued, defined, assumptions were made, identified and defended (this is where the outcome of many studies become manipulated) and testing then were to be conducted to validate the theoretical outcomes of such studies. I have seen, first hand, how almost any study can be distorted to achieve a desired outcome rather than a factual one. So I view much of this with a great deal of skepticism. Apparently not all sensors or sensor pixels are created equal. This is a variable that has to be scientifically accounted for along with many others. What I see is a very good product driven by competition to keep getting better. Don't ask me to define better because I don't know what form "better" will take. This has been a good and interesting discussion and I am sure the truth lies in some of these posts but for the life of me I don't know which ones they are. If it were in my former job I would just tell them to go back and do more work because I am not convinced.

[scott_ferris]

<p>Dick,</p>

<p>So it is your fault I have to work around GPS inaccuracies all the time? :-) In my proper grown up job I rely on GPS (well can't) several times a day.</p>

<p>Take care, Scott.</p>

[paul_russell1]

<p>Moores law kind of already applies in a physical sense, bayer pattern colour sampling and processing? <br>

Rather than upping the pixel count why not up the colour count? That is a 3x image device as used in professional camcorders (which have more in common with DSLRs insides than film cameras)<br>

This opens the door to neat tricks like pixel shifting to increase the perception of the appearance of detail?<br>

Shallow well ccds with proper HAD style micro lenses?<br>

Switchable FIT chips to avoid rolling shutter?<br>

Less noise.<br>

We might not need more megapixels, but extra picture information is always desirable, at the moment our cameras see on avarage a third of the colour that a film camera would see, thats before you take into account the higher dynamic range of some films.<br>

No more megapixels. Lots more colour please.</p>

[alan_rockwood]

<p>Let’s do a little calculation on diffraction limited optics and see what that implies with respect to sensor megapixels and lens performance. According to the Rayleigh criterion two spots at the film plane can be just resolved according to the following relationship:</p>

<p>X=1.22*L*F</p>

<p>where X is the distance between the points at the film plane, L is the wavelength of light, and F is the f-number of a perfect lens. Let us do the calculation for an f/2 lens, and assume the wavelength is 0.00055mm, which is in the green part of the visible spectrum.</p>

<p>From the above parameters we calculate</p>

<p>X=0.001342mm.</p>

<p>Let us take this number as the minimum pixel size required for a digital sensor to resolve two point objects and see how many pixels of this size will fit on a 24mm X 36mm full frame sensor.</p>

<p>The number works out to be 479.7 megapixels.</p>

<p>But wait! If you are using a Bayer sensor you need to double that number, which works out to 959.5 megapixel sensor. So you would need a 959.5 megapixel sensor to fully resolve a green image using a diffraction limited f/2 lens.</p>

<p>However, although it would be OK for the green part of the spectrum, because of the way the color sensors are interleaved on a Bayer sensor even a 959.5 megapixel Bayer sensor would be unable fully resolve the red or blue images, so even this estimate is a little optimistic.</p>

<p>Of course, you can’t go to the camera store and buy an f/2 diffraction limited lens, nor can you buy a 959.5 megapixel Bayer sensor camera. However, the point is that the present state of the art in consumer and professional digital cameras is not much better than two orders of magnitude worse than the fundamental optical limit. Therefore, one can conclude that there is still plenty room for improvement before we start bumping up against the limits, both with respect to lens performance and digital sensor geometry.</p>

[DickArnold]

Scott. If you are in an airplane you have the accuracy 13 feet vertically to make a precision landing, the integrity to know when you don't have that accuracy within 7 seconds, and the availability 99.999 to conduct aviation operations to, as I remember, about 4000 airports in the US. You need an aviation receiver that receives the Wide Area Augmentation signal provided by two geo stationary satellites and multiple ground stations. Ground based corrections are provided through the geos to aircraft in flight. It's been a while since I worked with this and I am not up to date. It took a lot of very accurate research to put this together. If you have some other receiver this may not be the case. BTW WAAS is incorporated in many inexpensive commercial receivers. However, they don't have the integrity functions that certified aviation receivers have. Anyway, WAAS is in daily use today after we started it in the early 90s. You have to blame DoD, Scott, for the unaugmented signal, however.as they invented it. We just adopted it to civil aviation use. Wikipedia has a very good description of WAAS.

[scott_ferris]

<p>Dick,</p>

<p>I know, I was just being flippant :-), I did it in part to cool the thread, but who really cares about people shouting at each other about totally hypothetical nonsense over the internet? I work with DoD castrated marine GPS, on a good day I can get to within 100ft, if I ignore the altitude readings which normally put me between -20ft - +150ft when on the surface. On a more normal day 300ft is as accurate as I can get from my array of instruments, good old lineups and depth sounders doing the more accurate work.</p>

<p>Take care, Scott.</p>

[Landrum Kelly]

<blockquote>

<p>If resolution were everything, then just consider the implications back in the film days. Wouldn't we all be carrying around 8x10 plate cameras?</p>

</blockquote>

<p>Well, of course resolution has never been everything, Sarah, but what if we could get really great resolution <em>and</em> great portability? I personally like the idea of getting medium format resolution on 35mm sensors. It looks like we are getting very close to that. I find the idea very appealing.</p>

<p>Large format? Nah. Just my way of saying that there is no such thing as too much resolution, everything else being equal. </p>

<p>--Lannie</p>

[c._sharon]

<blockquote>Let’s do a little calculation on diffraction limited optics and see what that implies with respect to sensor megapixels and lens performance.</blockquote>

<p>Here's a link to link to an article that attempts to provide an answer to precisely that<br>

<a href="http://www.luminous-landscape.com/tutorials/resolution.shtml">http://www.luminous-landscape.com/tutorials/resolution.shtml</a></p>

<p>At the bottom of the article there is Table 3 which lists the megapixels for various sensor formats at different apertures for blue, green-yellow and red light wavelengths. For example, the author mentions that a 35mm sensor with a lens at f/11 will at best have a maximum resolution equivalent to 16 MP. But the maximum resolution in megapixels can be much higher at larger apertures.</p>

<p>Interestingly the table says that at f/16, one can get a maximum resolution of 3 MPs from an APS-C sensor and 7 MPS from a 35mm sensor. I point this out as there was a post earlier about often using small apertures like f/16.</p>

[kelly_flanigan1]

<p>RE" The question REALLY should be: How many MP do you want/need?"</p>

<p>A Rational person just gets a bigger displacement weed eater; lawnmower; chain saw and does not go nuts over Horsepower per cubic inch.</p>

<p>It is cheaper to just a bigger weedeater engine thatn get one with a turbocharger.</p>

<p>One can dream up all sorts of crazy high numbers with the typical amateur's "best caseing" a giant string of events' to "cook the books" ; to get high numbers/</p>

<p>******Look at dedicated film scanners; a declining market that Minolta, Canon and Nikon have dropped. Most all high end scanners peaked about 4000 dpi; a few were at 5400.. This peak occurred over 12 years ago when 4000 dpi scanners were a hot thing. The real only advancement has been increased bits and a slight noise drop. 4000 dpi was the to number for most all film scanners because that is all folks originals reall hold; on *PRACTICAL BASIS* of the sharpest originals. That is 4000dpi over a 24x36mm frame; ie 3780 x 5669 pixels; about 21 megapixels. Without the transfer function of film; one can go higher; but probably not too high.</p>

<p>In reference to a F2 Diffraction limited lens for a 35mm camera; over its 24x36mm frame. This will happen when Obama cuts taxes; or Unions ask for lower benefits; or TV's go back to using 12AU7 tubes.</p>

<p>On might as well mention how a teenager has 1 cellphone minute per month; or a golfer one needs one golfball for his entire lifetime; or a woman only needs one pair of shoes.</p>

<p>Extracting say 40 USEABLE megapixels out of a 24x36mm frame; means alll the best case things have to happen. Here I am doubtfull that in 10 years this will be easy or commonplace. Few today even reach the limits of the current high end dlsrs.</p>

<p>(1) That is OK with a lay amateur; they have no deadlines; no clients; they can spend hours waiting for the perfect pitch.</p>

<p>(2) an actual pro has to have some design margin in a shot. That is why a pro would use a MF back for 40 megapixels TODAY ; and not fart around "hoping" that it will work with 35mm dslr.</p>

<p>In shooting giant artwork here I have already run into applications ten years ago were my 35 and 50 megapixel scan backs are not enough to pull out tiny features; BUT they were good enough for the clients needs. ie I do not have enough work to justify buying a 100 megapixel scan back.</p>

<p>In some fine county maps last week; they had types so fine that instead of usings a typical 300 dpi setting; a 600 was used with my 36" wide RGB color scanner.<br>

A 500 setting left out some subtle features in the clients absurd tiny fonts from hell.<br>

Thus a 36x48" map at 600 dpi scan is a monster 21600 by 28800 pixel scan; ie 1.74 gig file; a 622 megapixel file. The best 35mm dlsr on the planet is a childs toy for shooting such a map.</p>

[Landrum Kelly]

<blockquote>

<p>That is OK with a lay amateur; they have no deadlines; no clients; they can spend hours waiting for the perfect pitch.</p>

</blockquote>

<p>Well, Kelly, I have no photographic deadlines to meet, but I do have deadlines and other obligations and aspirations in my "real" job. For me, not having to scan means a lot, and a "child's toy" that gets me forty or so megapixels will likely have to do on my budget.</p>

<p>Sure, if I had worlds enough and time (and money), I could get the best of everything for all possible applications, but, since I don't have any of those, it would be nice to have something that is (1) portable, (2) affordable, (3) better than anything out there now in 36 x 24 format, and (4) capable of using the lenses I have now, while still being compatible with upgrades that I might (or might not) be able to afford in the future.</p>

<p>That much could be around the corner with Canon's next flagship offering--and the consumer grade counterpart that is sure to follow. In the meantime I am enjoying immensely the 5D II with a very few pieces of L glass, another compromise that works pretty well with my budget and my time constraints.</p>

<p>We all have to operate within the parameters of our actual lives, with all of our other aspirations and obligations. Everybody's situation is different. I have books to write, courses to prepare for, a new language or two I really want (but don't have) to learn before I die, etc. I have obligation and aspirations, that is, and some of what I want is not obligatory but is desirable to me for my own personal reasons. When I get to go out to shoot, I want to get good shots, but the last thing that I want is to have to clone the dust off of scans. I hate scanning. If I did not, I might be out shooting a Hasselblad this afternoon. Instead, I think that I will study a bit of French, then go out with my 5D II later and see what I can get with the convenient and familiar cameras that I have--until I can go up a notch without breaking the bank.</p>

<p>The challenge, "But do you really <em>need</em> that?" is ultimately a very imperial kind of question to ask. I need very little of what I have and what I do. What I want to do with my life is very much my business, in any case, and I am tired of someone else telling me, "<em>x</em> ought to be enough for anybody." They can speak for themselves. I have my own opinions and valuations of what would fit into my life--and my larger life plan.</p>

<p>--Lannie</p>

[alan_rockwood]

<p>C. Sharon, Interesting about the f/11 needing about 16 megapixels. (I assume they are talking about real megapixels, not Bayer megapixels.)</p>

<p>If you scale the requirement at f/11 to estimate the requirement at f/2 you get 484 megapixels (i.e. a 968 megapixel Bayer sensor). That is almost exactly the same answer I got.</p>

<p>Scaling law for this calculation is (11/2)^2.</p>

[c._sharon]

<p>Alan, The megapixels listed in table 3 of the article are for Bayer sensors. These values are computed based on the pixel pitch required to resolve a line pair for Bayer sensors as listed previously in table 2. Also, the megapixels in table 3 are what you can get from the sensor.So for 35mm sensor at f2 it is 471 MPs, not 471x2. And these are real megapixels.</p>

<p>So to answer the OP, for a 35mm Bayer sensor the upper bound on the resolution is 1885 MPs, depending upon the aperture. For example it is 60 MP at f/5.6, considering green-yellow light as the human eye is most sensitive to it . It is less at a smaller aperture and more at a larger aperture, all the way to f/1.</p>

[swilson]

<p>

<p>My take on this is that going to 40 MP for a FF sensor would not be hard at all, no new lenses would be needed. My 1.6 crop camera makes good use of a number of my lenses including my 50mm lens. What limits 35mm film photography is not the lenses as much as the film itself, put in a high resolution sensor and you get more detail. If this were not the case then teleconverters would not work, a teleconverter is only useful if the lens it is used with has more resolution then the film or sensor can capture.<br>

The question is would putting 40MP on a camera make it more or less useful then a 20MP camera, which of course depends on what you are doing with the photos. I have made 12x18 inch prints from my supper sharp stitched photos and from single frame from a 8 MP DSLR, people don’t look close enough to be able to tell the difference. There are two reasons why I sometime want more resolution, being able to crop more and using the photo to document something. I like to do before and after photos of places, separated by years of time and often capturing more detail makes it more fun when comparing the photos.</p>

<br>

But over all for most uses 20MP is more then what people need, but if someone comes out with a 40MP camera there will be a market for it.</p>

[matthias_meixner2]

<p>When it comes to image quality resolution is only one aspect of the game. Sure, cramming more pixels on the chip increases resolution but at the same time reduces sensitivity and increases noise. So the real question is not how many pixels fit on a chip but the question for the best compromise of resolution, sensitivity, noise and dynamic range. Hopefully marketing will not sacrifice the other three for resolution.</p>

[keith reeder]

<blockquote>

<p>Sure, cramming more pixels on the chip increases resolution but at the same time reduces sensitivity <strong><em>and increases noise</em></strong>.</p>

</blockquote>

<p>Oh, for the love of God, how many more times with this nonsense?</p>

[craigd]

<blockquote>

<p>Oh, for the love of God, how many more times with this nonsense?</p>

</blockquote>

<p>Funny, I was just thinking the same of you. I notice that you are not replying to any of the people who have shot down your previous comments. You just come back with the same line, apparently not having learned anything.</p>

[kelly_flanigan1]

<p>40 Megapixels in full frame 24x36mm is a 5165 by 7735 pixel image; ie 40,0029,250</p>

<p>. That is a 5466 pixel per inch sensor. ; 215.2 pixels per mm; 4.65 micron sensor.</p>

<p>All one needs is a lens that resolves 108 line pairs per mm corner to corner; any kit zoom should do this in the lay persons Harry Potter world of make believe.</p>

<p>All you are are asking for is 100 + line pairs per mm in the far corners of a 24 x36mm frame; some thing that has never happened yet.</p>

<p>This is why folks love marketing BS; folks are "consumers"; that new gizmo will be better!:)</p>

<p>I can take all the pro and amateur image I have ever scanned since 1989 and non really have 5466 dpi worth of info in the corners; this is beyond the off axis performance of a full frame lens.</p>

<p>Thus a 40 Megapixel Full Frame dslr might use the central core's vast details; but the far corners will be wasted. Thus does marketing make a camera for bragging rights; that one practically cannot use in the corners?</p>

<p>Thus an actual pro would prefer to use a MF 40 megapixel camera; one that actually has to design margin and works.</p>