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

[kelly_flanigan1]

<p>The Shift Nikkor 35mm F2.8 PC can shift 11mm; thus it covers somewhat more than 24x36mm. On a 4x5" camera; one gets a round image like an 1890's Kodak!<br>

<img src="http://i4.photobucket.com/albums/y148/ektar/Speed%20Gr%2035mmPC/tripods-298.jpg" alt="" /></p>

[alan_rockwood]

<p>Kelly,</p>

<p>Why the excessive interest in the corners? Don't forget the center!</p>

[kelly_flanigan1]

<p>Alan;</p>

<p>Because the clients stuff I shoot actually has details *they* want recorded in the corners.</p>

<p>It is an actual requirement.</p>

<p>One could tell the client to back sand; or say one only has a wimpy dlsr that cannot pull out the corner details.</p>

<p>Or one can shoot a mess of panels and stitch them together too. I have stitched stuff together even with several 35 or 50 megapixel panels.</p>

<p>A straight on shot is easier; whether dslr or scan back.</p>

<p>If I used a 40 megapixel dslr that only was like a 12 at the corners; it would not record the required details on may jobs. If you have no clients that have requirements; one can ignore the corners.</p>

<p>Thus here a real 35 or 50 megapixel back and lens combo is used with recording larger pieces of art; to capture the details a client wants.</p>

<p>The phase one scan backs I use work like this:</p>

<p>http://www.betterlight.com/how_they_work.html</p>

<p>I use them sometimes to shoot buildings or a landscape with a laptop. Both the Betterlight and PhaseOne's are sub 4x5"; Betterlight is 72x96mm; Phase ones here are 70x100mm.</p>

<p>With the growth of the dslr over the last 17 years; some folks use them today for smaller artwork pieces; I do many times too where the artwork is smaller or has less details.</p>

<p>If one considers the area of a 24mm round image; it is 452mm squared. A 24x36mm film frame has an area of 864mm squared. If a lens has a "40 megapixel class performance over a 24mm circle; the film frames height; 48 percent is at the outside. The the corners hold a lot of real estate</p>

<p>The scaling of tiny sensors to big 24x36mm sensors has its modeling/scale failings. It is like going from ants to elephants.:)<br>

From an amateurs standpoint; corners probably do not matter; bragging about megapixels does. Thus if Canon makes a 40 megapixel FF dlsr; folks will buy it even if the corners and edges are just like a 12 or 18 camera. Since there is no client or goals; a poor corner or edge doesnt matter.</p>

[danielleetaylor]

<p><em>If you downsize a 21MP image to 12MP you get two 12MP images with similar noise, as Keith says.</em><br>

<em>If you leave the 21MP image at 21MP you get a higher res image with more noise, which is the trade-off Sarah points out. There are others too, such as more processing time and fewer images per card.</em><br>

<em> Which leads one to ask, what is the point of Keith's method?</em></p>

<p>Images are printed to a specific size. Which means when you print a 12 MP and a 21 MP image each to 16x24, you are "downsizing" the 21 MP image relative to the 12 MP one. The noise in the 21 MP image, indeed all pixel level artifacts, are magnified less relative to the 12 MP one.</p>

<p>Pixel peeping (i.e. viewing images at 100% in Photoshop) gives a distorted view of noise performance because higher resolution images are magnified more. It therefore leads people to falsely conclude that sensors with lower pixel density also have lower noise. Though there is a point where increasing pixel density will adversely affect noise performance, DSLRs have not yet reached that point. At the densities found on DSLR sensors, for a given level of technology total sensor surface area determines image noise, not pixel density.</p>

<p>I would happily post sample after sample to prove this point, but I would have to take crops from DPReview and Imaging Resource test samples, and photo.net will not allow those to be posted.</p>

[alan_rockwood]

<p>Kelly, If I can read between the lines and risk an interpretation, I think what I hear you saying in your last post is not that that there is necessarily something wrong with the hypothetical 40 megapixel sensor (if it could actually be built), but rather that you don't think there can be lenses that are sharp enough in the corners to take advantage of a high pixel count. In other respects it sounds like you are, in fact, in favor of high pixel count sensors, just not in a 24x36mm format.</p>

<p>So let me pose a very hypothetical question: If there were a 24x36mm format camera that had a 40 megapixel sensor, and if there were lenses that were sharp enough in the corners to meet your needs, and if it were affordable within your budget, would you find such a camera potentially useful?</p>

[Landrum Kelly]

<blockquote>

<p>Pixel peeping (i.e. viewing images at 100% in Photoshop) gives a distorted view of noise performance because higher resolution images are magnified more. It therefore leads people to falsely conclude that sensors with lower pixel density also have lower noise. Though there is a point where increasing pixel density will adversely affect noise performance, DSLRs have not yet reached that point.</p>

 

</blockquote>

<p>Daniel, no matter how many times or how many ways you say it, some people are just not going to get it. I am not trying to sound elitist in the least, since there is a lot that you and Kelly and others get that I do not, but this is a truly essential point that you are making.</p>

<p>Some people need to print, I think. Screen comparisons are only going to carry one so far.</p>

<p>--Lannie</p>

[Landrum Kelly]

<blockquote>

<p>If there were a 24x36mm format camera that had a 40 megapixel sensor, and if there were lenses that were sharp enough in the corners to meet your needs, and if it were affordable within your budget, would you find such a camera potentially useful?</p>

</blockquote>

<p>I dare not try to answer for Kelly, Alan, since he knows at least a thousand times more than I do, but I think that there is another problem besides lenses: the corners are going to have light drop off on DSLRs because digital sensors perform best when they are perpendicular to the light. Film (as I understand it) does not have that particular problem.</p>

<p>Rather than offering an answer, however, I am left with a question: how does one get around that particular problem in the corners? Better resolving lenses surely would not solve that problem for DSLRs. A medium format sensor and lenses with larger image circles would seem to be more promising in terms of reducing light fall-off in the corners than would 36 x 24. At least, so it seems to me.</p>

<p>Somebody please correct me if I am wrong on these points.</p>

<p>--Lannie</p>

[craigd]

<p>Dan South:</p>

<blockquote>

<p>Any of the current tilt shift lenses would cover 30x36 without any vignetting and would even have room for generous movements. Canon makes four of these lenses...</p>

 

</blockquote>

<p>Right, that's four manual-focus-only lenses out of Canon's whole EF lineup, the two newest models costing more than $2000 apiece. Not exactly typical, are they?</p>

<p>If someone proposed a minimum speed limit of 190 mph for freeways, and someone else argued that most existing cars can't go that fast, would it really be a sensible refutation to point out that a Lamborghini Diablo can?</p>

[dave_redmann]

<p><em>Pixel peeping (i.e. viewing images at 100% in Photoshop) gives a distorted view of noise performance because higher resolution images are magnified more. It therefore leads people to falsely conclude that sensors with lower pixel density also have lower noise. Though there is a point where increasing pixel density will adversely affect noise performance, DSLRs have not yet reached that point. . . . Daniel, no matter how many times or how many ways you say it, some people are just not going to get it.</em></p>

<p>I'll omit discussing the basic-but-critical question of whether the photon-receptor-area to other-sensor-area ratio is equal for all sensor resolutions.</p>

<p>First of all, magnification is not a function of pixel density; it is a function of sensor size and print size. Second, by your statement, "higher resolution images are magnified more", I assume you mean that, for a given print size, higher resolution images are upsampled less or downsampled more than an otherwise-comparable lower-resolution image. In other words, if you want to print an 8x12-inch print at 300 ppi (2400 x 3600 pixels), with a 24 MP camera (4000 x 6000 pixels), each pixel in the print is (roughly, sort of, to simplify things) the average of 2.78 pixels on the sensor, and because the noise is <strong>presumably</strong> random and the signal is presumably not random, the (in effect) averging reduces the noise relative to the signal; but if we have a 6 MP camera (2000 x 3000 pixels), each pixel on the sensor will have to provide 1.44 pixels on the print, and in effect noise is magnified. (Put another way, each sensor pixel accounts for four times as much print information with the 6 MP camera as the 24 MP camera, making per-pixel noise much more critical.) (This is a considerable oversimplification, but I think it mostly suffices for present purposes.) Basically, the concept is that if you measure the same thing more than once and average your measurements, the average will probably be closer to the true quantity that would at least the majority of single measurements.</p>

<p>But where this may break down is the assumption that noise is random pixel to pixel. No less a noted voice than Phil Askey of DPReview has taken issue with this very point in his blog entry, "Downsampling to reduce noise, but by how much?". See <a href="http://blog.dpreview.com/editorial/2008/11/downsampling-to.html">http://blog.dpreview.com/editorial/2008/11/downsampling-to.html</a>. There he explains, "One of the reasons that theories about downsampling reducing noise don’t appear to work in practice is that the theory assumes noise is random. Unfortunately, this isn’t necessarily true. Noise at a single photosite will effect adjacent pixels as part of the demosaicing process. So noise doesn’t occur as individual pixels but as grain. The mathematical theory may tell you that downsampling works but it won’t if your noise grains are any larger than one pixel (and they nearly always are from a camera with a bayer color filter array)."</p>

<p>So while I think I understand your argument, and think there is some truth to it, it isn't as simple as you appear to want to make it seem, and in any particular case (5Ds Mk. II vs. D700 etc.), it might or might not suffice to carry the day. That is, the larger-resolution sensor provides some advantage through more downsizing or less upsizing, but it's not as big an advantage as the basic math of random noise implies, and in comparing any two cameras, there does not appear to be a simple mathematical relationship among their pixel counts, per-sensor-pixel noise levels, and per-final-output-pixel (print) noise levels.</p>

[karl_lehmann]

<p><em>"Images are printed to a specific size. Which means when you print a 12 MP and a 21 MP image each to 16x24, you are "downsizing" the 21 MP image relative to the 12 MP one. The noise in the 21 MP image, indeed all pixel level artifacts, are magnified less relative to the 12 MP one. </em>- <a href="../photodb/user?user_id=2129602">Daniel Lee Taylor</a></p>

<p>This is true, but it misses the point<em>.</em> The 21MP image gives you the option of printing at a larger size and/or cropping more. If you do so, you will see more noise. At some pixel density, with a given level of technology, the noise will reach a point where you will no longer wish to print at larger sizes. That is the trade-off.<em> </em></p>

<p><em>Pixel peeping (i.e. viewing images at 100% in Photoshop) gives a distorted view of noise performance because higher resolution images are magnified more.</em></p>

<p>That is only true if you misinterpret what you are looking at.</p>

<p>Daniel, it's obvious that you understand all this so it's silly to keep debating it.</p>

 

[danielleetaylor]

<p><em>This is true, but it misses the point. The 21MP image gives you the option of printing at a larger size and/or cropping more. If you do so, you will see more noise. At some pixel density, with a given level of technology, the noise will reach a point where you will no longer wish to print at larger sizes. That is the trade-off.</em></p>

<p>I would agree with this, though I don't think we're near that density yet with DSLRs. P&S bodies have reached and passed that density, which is why Canon has scaled back some models.</p>

<p><em>Daniel, it's obvious that you understand all this so it's silly to keep debating it.</em></p>

<p>I was gone for a while and jumped into the thread where I left off. Responded to one post, read on, and then realized the thread was moving away from that subtopic. Sorry for beating a dead horse.</p>

[sarah_fox]

<p>Karl, you wrote (regarding camera vibration):</p>

 

<blockquote>

<p>I can answer that because I often stack images to reduce noise and/or increase dynamic range. If I use MLU and a remote, sequential images at 21MP are in register to within less than 1/2 a pixel (usually much less). That means at 100MP they ought to line up to within a pixel or better.</p>

 

</blockquote>

<p>Well, that's repeatability, not total vibrational excursion. If your repeatability is on the order of 1/2 pixel at 21MP, I suspect total vibration is more, and that you simply have a repeatable blur pattern that might extend over a pixel or so (just a guess). This would still beg the question of how much a camera flops about during an exposure. I'm sure the answer depends greatly on shutter speed, and I suspect only laser vibrometry would give us any sort of meaningful data. (I used to have a laser vibrometer in my lab, but alas, I'm now an academic sector dropout.)</p>

<p>Anyway, taking your estimate of 1/2 pixel at face value, I would say the 21MP resolution is already the limit that can be achieved with the rigidity of your system. Image instability on the order of 1 pixel would suggest to me that there are excessive pixels and that the image could be just as well represented with half the pixel density. Of course with a long enough exposure, shutter vibration effects are minimized, and higher resolution may still be achievable.</p>

<p>I'm not saying this problem can't be overcome. A beefier mount might enable a person to achieve the necessary stability to take full advantage of a 100MP sensor, but I suspect there would have to be a redesign of the conventional 1/4-20 threaded blind-hole. (I've thought of making a beefy frame that would extend from tripod thread to hot shoe, gripping both the top and the bottom of the camera.) Although this is a fixable problem, it's not a problem that most photographers can appreciate or care about. Therefore I suspect it will never be fixed, even if someone makes a 400 MP DSLR. As you suggested, such cameras will be marketed to the well-heeled vacation snap-shooters who buy most of the cameras.</p>

 

[mike_hitchen]

<blockquote>

<p>...however, I am left with a question: how does one get around that particular problem in the corners? Better resolving lenses surely would not solve that problem for DSLRs.</p>

</blockquote>

<p>Use of curved sensors? These have been developed.<br>

In addition, lenses would not need to be rectilinear (or less so, anyway) so mean smaller and lighter lenses. If you are going to totally redesign the camera system why not do it properly? </p>

[kelly_flanigan1]

<p>Alan; RE :"</p>

 

<p>So let me pose a very hypothetical question: If there were a 24x36mm format camera that had a 40 megapixel sensor, and if there were lenses that were sharp enough in the corners to meet your needs, and if it were affordable within your budget, would you find such a camera potentially useful?</p>

 

<p>"<br>

*YES* I would find it usefull to shoot framed artwork where It will not go through my 36" wide RGB scanner (HAS LIMIT on thickness) and it would be radically quicker than my old tethered scan backs.</p>

<p>It is not that I oppose making a 40 Mpixel dslr; it is having a healthy concern whether there is real corner and edge performance.</p>

[kelly_flanigan1]

<p>I purposely have not mentioned the issue of noise; since the subject is already quite confusing and filled with folks on both sides of the fence.</p>

<p>The dumb slow as dirt scan back's of mine have very low noise; but have giant pixels on the sensor bar. The 35 Megapixel scanners equalivalent pixel size is a giant 14 microns.</p>

<p>Since technology changes it is hard to debate noise issues. Two P&S or dlsrs with say 6 micron pixel sites can have different noise figures; because of the way they are driven and the technology used. </p>

[g dan mitchell]

<p>One thought, perhaps not closely connected to the most recent messages in this thread... When it comes to the relationship between sensor resolution and lens resolution, there are three logical possibilities. I'll list them below with a brief comment on each. My preference should be clear by the end.</p>

<ol>

<li><em>Lens resolution exceeds sensor resolution</em>. Given this situation, one we are familiar with from cameras of the past decade, most photographers who are in a position to see benefits from increased resolution in their photographs would be happy to see sensor resolution increase.</li>

<li><em>Lens and sensor resolution are equal</em>. I think this could sound like the ideal, but if you think about this you'll soon realize that it is impossible or at least quite difficult to quantify. Equal to what? The best portion of the best lens you use at its sharpest aperture in the center of the frame? Equal to the average resolution of all of your lenses? </li>

<li><em>Sensor resolution exceeds lens resolution</em>. In this case, the system has enough reserve to make use of whatever resolution your lenses can deliver in the most optimum situation - in other words, sensor resolution ceases to be a limit. </li>

</ol>

<p>Of these three, the third option seems best to me as long as it can be accomplished without sacrificing other aspects of image quality and as long as the cost is manageable. It is certainly true that most photographers will see no real world benefit from this situation - but some will. If it can eventually be achieved at a reasonable cost and with no other downsides... why not?</p>

<p>Dan</p>

[kelly_flanigan1]

<p>Dan;<br /> One can get case (3) today with a premium disposable 35mm camera.</p>

<p>Here I get law enforcement crime scene images like that. The central core is so sharp that a 4000 dpi image shows more info than a 2720 dpi one. A giant 30x42 or 36x48" poster for court has a tack sharp center area and poorer corners; a sort of weird effect.</p>

<p>At first I thought I had some oddball scanner issue. One has a Leica Summicron like image in the center; and the corners worse than my 50mm F2.8 Domiplan on my old Exakta; ie instamatic like.</p>

<p>I wonder is folks who buy a 40 or 50 megapixel high end FF 35mm dslr are "OK" with prints that have a sharpness response that varies across the image?</p>

<p>The "downsides" is somebody might expect that one has a useable 40 or 50 megapixels worth of info. It would be like buying a pickup truck that gets 50 MPG at 40 MPH; and the same 25MPG at 70MPH; some folks might be disapointed.</p>

<p>If one shoots birds; maybe case (3) would be a happy thing; the bird is in the center.</p>

<p>If one shoots maps and artwork with details and ones rigs at the corners is really like an 18 megapixel rig; one would be either loosing details; or resorting to stitching together two shots to give a real 40 megapixel performance across the entire map.</p>

<p>If my current 35 and 50 megapixel scan rigs had soft corners and was case (3); then there is real no reason to shoot a map in these high res modes anyway. I might as well reduce the resolution setting so the response is uniform; then shoot several panels say 4 and combine them to reproduce a map properly. The whole reason some of us use a big high resolution settup to have a full response corner to corner; to get away from paneling; ie reduce labor. Thus case (3) has really no place in mapping; ie whussy bad corners due to the lens. It is like replacing a fine 6 element enlarging lens with a 1950's triplet; we are back to a starter setup with weak corners; something my brothers Sears Testrite MF enlarger had with its perflex lens!</p>

<p>Those who might like case (3) might have non uniform subject matter; you focus on a bird; ballplayer; portraits; and it is in the central core of the 24x36mm frame. If one wants a uniform performance then case (1) is wanted; ie for mapping and shooting fine artwork with corner details.</p>

<p>Here I actually use a dslr to shoot smaller pieces of artwork; thus a uniform response would be wanted; not just a sharp central core and weak corners.</p>

<p>Many camera/lens combos today and even back to photographys beginnings have a response already like case (3); this goes back to the pre civil war era and glass plates too.<br>

One is already getting case (3) already with current sensors and films with most folks images; they always do no shoot at F11!</p>

[dan_south]

Craig, the point is that those lenses are available today. Add one good midrange zoom, a tele, and a macro and Canon is

all set for larger than full frame sensors. If you think canon's t/s lenses are cost prohibitive check out the lenses for the

Phase One and Leica S2 systems. And what's wrong with manual focus? These days I use live view frequently, much of

the time with a t/s lens. The results are almost always better than spraying and praying with autofocus and dealing with a

lot of perspective distortion.

[karl_lehmann]

<p>Sarah, you wrote,<em> "Well, that's repeatability, not total vibrational excursion."</em><br /> <em> </em><br>

<br /> I think it's some combination of the two because the sequential exposures are usually made at different shutter speeds. The largest excursion occurs near the beginning of the exposure, so if it's significant the exposures won't line up. In fact you can almost eliminate the effect by using long exposures - with a 10 second exposure you don't even need MLU. With very short exposures (less than 1/60 or so but it depends on your system) total vibrational excursion will be more important. But camera-induced vibrations will not be the limiting factor in useful pixel density because you can reduce them arbitrarily by increasing exposure time.</p>

<p><em>"taking your estimate of 1/2 pixel at face value, I would say the 21MP resolution is already the limit that can be achieved with the rigidity of your system"</em></p>

<p>But that is not my estimate, with good technique it is usually much less. The reason I mentioned 1/2 pixel is because that is what I look for - if the difference is greater than 1/2 pixel the images need to be aligned before stacking. I don't like to stack when it's that sloppy because I am trading resolution for increased dynamic range and reduced noise. But a slight reduction in resolution is far better than blown highlights and blocked up or noisy shadows.</p>

<p>Usually the difference is not even detectable unless I have touched the camera between exposures (which I would consider poor technique, but sometimes you have to do it). Based on that I would say 21MP resolution is nowhere near the vibration-based limit. And my system is nowhere near the ultimate possible. Given a 100MP sensor I might have the incentive to buy a heavier tripod or lengthen my exposures, for example.</p>

<p>The bottom line is, we will reach limits based on diffraction (or, possibly, signal-to-noise ratio) long before we run out of ways to reduce camera movement.</p>

[daniel flather]

<p>Who cares.</p>

[Landrum Kelly]

<p>Daniel, I am going to dignify your flippant comment with an honest response.</p>

<p>I care, and I care for a very specific reason. I have aspirations in the realm of medium format photography--but I don't like to scan film. Therefore I am considering either a medium format digital back or else starting with a new digital medium format system.</p>

<p>Before I make the leap, however, I want to examine the possibility of getting true medium format quality out of a 36 x 24 sensor. Is it possible? Will it ever be possible? What would be the downside? What unanticipated additional problems and/or costs might I encounter if I made the leap to medium format digital--as opposed to staying with the Canon EOS system?</p>

<p>These kinds of questions were behind my posting of the question on this thread in the first place.</p>

<p>I really could not afford both Canon EOS and medium format digital, and so I would have to sell all or most of my Canon gear to make the transition. I am talking about a potential major transition which could be quite costly. My first love really is landscape photography, however, and I would especially like to do some serious work in the mountains of western North Carolina, which start in earnest about an hour and a half from where I live. (The foothills--the Brushy Mountains--are less than an hour away from my house.)</p>

<p>I have also considered the stitching option for such work, but so far I am unconvinced that that option would give me the results that I want.</p>

<p>--Lannie</p>

[Landrum Kelly]

<p>One of the alternatives to the 36 x 24 sensor of the Canon EOS system that I am strongly considering is the Hasselblad back for the 6 x 6 cm cameras:</p>

<p>http://www.hasselbladusa.com/news/hasselblad-debuts-cfv-39-digital-back.aspx</p>

<p>This back has a sensor that is just about exactly twice the size of the 36 x 24 mm sensor of the Canon EOS system. It is slower, of course, but that would be expected and would not be a real problem with the kind of work that I am anticipating.</p>

<p>One advantage is that I already have the Hasselblad bodies, lenses, and backs. Most of the expense would go into the back, of course, but the adapter does not come cheap at B&H.</p>

<p>--Lannie</p>

[Landrum Kelly]

<p>The only other alternative that I am seriously considering at the moment is the new Pentax 645D, already out in Japan but currently unavailable in the United States. Its primary advantage over the Hasselblad would be cost:</p>

<p>http://www.luminous-landscape.com/reviews/cameras/pentax645d-1st.shtml</p>

<p>http://www.dpreview.com/news/1003/10031002pentax645d.asp</p>

<p>The future of Canon EOS full-frame digital is thus very relevant to me in terms of what I would like to do. I understand that the 36 x 24 sensor will never match true medium format, but I cannot help but wonder just how good it might become in the near future.</p>

<p>Thus my original question:</p>

<h1>How many usable megapixels will go on a full-frame sensor?</h1>

<p><br /> For me the question had specific relevance to changes that I am presently contemplating, as noted above.</p>

<p>--Lannie</p>

[dcstep]

<p>Lannie, my first camera was a 44x44mm TLR, so I know what attracts you to medium format.<br>

Given you love of scenics, it makes a lot of sense to consider the Hasselblad and, maybe, the Mamiya. Consider the whole system, particularly the lens choices for scenic work.<br>

Also, if you don't already have one, rent a Canon 5D2 and a 17mm TS-E lens and one of their L-series wide zooms, a good sturdy tripod and head for the hills. Take a few images at low ISOs, small apertures, with the mirror locked up and with cable release. I think you'll be very pleased. I DO think there is more to come out of FF sensors, but they're very good at this point.</p>

<p>Dave</p>

[Landrum Kelly]

<p>Thanks, David. I do have the 5D II and a good Gitzo pod with an AS Z-series ballhead. I just sold my original EF 24mm f/1.4 (not the Mark II) because I knew that it was not going to do what I have in mind. I will have to consider other EF lenses before I make the leap, of course.</p>

<p>This is not a switch that one makes without substantial deliberation.</p>

<p>I have to say that I am pretty impressed with the 5D II at the lowest possible ISO.</p>

<p>--Lannie</p>