Maximum pixels

<p>With Sigma's release of its newest product with a whopping 43 megapixels sensor, I was curious with two related topics: 1. At what point, if ever, is "maximum" megapixels just a hype selling tool? 2. Are today's lenses capable of allowing the clarity of the shot to be truly captured by the sensor? In other words, is there not a maximum megapixel capability of the lens?</p>

<p>The Sigma SD1 (I assume that's the one you mean) doesn't have a 43 MP sensor. That's just marketing hype. It has 14 MP with separate photosites for each color in every pixel, in contrast to the more common type of sensor that detects only one color per pixel and then uses a demosaicing algorithm to interpolate colors. This does not magically add up to 43 MP, which would imply resolution superior to some medium format digital systems (e.g. Pentax 645D, Leica S2). The images that come out of the camera are 14 MP, not 43 MP. However, given comparable lenses, one would expect/hope that the SD1 would produce finer detail than a 14 MP camera with a Bayer array and an anti-aliasing filter.</p>

<p>The most sensible way to look at this is that the SD1 is potentially capable of full 14 MP resolution, while a 14 MP camera with a conventional sensor, although it does produce 14 MP files, has somewhat less real resolution (how much is debatable, but probably around 10 MP) due to the blurring effects of de-mosaicing and anti-aliasing.</p>

<p>Not having actually used the SD1, I don't know if it is capable of out-resolving today's best lenses, but this brings up another problem: today's best lenses are mostly not from Sigma, and the SD1 isn't mount-compatible with any other manufacturer. So I have to wonder how much the extra resolution of the Foveon sensor really matters if you have to shoot with inferior lenses.</p>

<p>Phase One had a 50 megapixel 4x5 scan back 12 years ago. It scans a 7x10cm area and the effective sensor/bar pitch is so large that a 1950's lens often is overkill.</p>

<p>Veronica does bring in a significant point: sensor size. 50 megapixels on an APS-C sized sensor, or 50 megapixels on medium format does make a lot of difference with regards to your second question.</p>

<p>Point 1 also comes down to size - at what size do you print? I've got a 40x60 centimetre print (from a normal lab), from a 6MP Nikon D50 with a mediocre lens. Looks perfectly fine. But I don't think one should expect much more than that size from the 3008*2000 resolution. For the print sizes most people use, 10 to 12 megapixels will typically suffice. Of course, it is nice to have some room to crop, so the extra resolution can still be useful.</p>

<p>Point 2 - there is a point where lenses become the limiting factor. The 16 and 18 MP APS-C sensors are already pushing the limit on quite a lot of lenses and revealing flaws that lower-pixel count cameras weren't. But this, as said above, includes also the sensor size - so one should look here purely at the physical size of each pixel. One cannot say "xx amount of pixels" is the maximum of this lens. It's more along the lines of "pixels per square mm sensor". For really good lenses, it's not much of an issue (yet), though.</p>

<p>SD1 actually has the large number of pixel sensors separated into 3 color planes.<br>

It really has the 43+ millions discrete measaured values, that all are real and contribute to dynamics and quality of the final picture.</p>

<p>If you are on quest to degrade the Sigma SD1 camera, go ahead and see what really happens in cameras that utilize Bayer filters. Apply your reasoning to a Bayer filter camera.<br>

E.g. from a 12 MP DSLR camera Bayer filter separates the 12 million lixels into 3 colors, and what you really get ? 12/3 = 4 million pixels camera, - by the same measure that you are trying to apply to SD1. Then "magically" from 3 x 4 MP, you get 12 MP. Similarly SD1 "magically" can join 3 x 14 MP to get 43+ MP.</p>

<p> </p>

<p><em>It really has the 43+ millions discrete measaured values</em></p>

<p>Of course, but since the photosensitive elements for the three colors are on top of each other, it all yields a 15 million pixel image (pixels are elements of the digital image, photosites are components on the sensor). There is no basis for the claim of 46 MP since all the information from the sensor can be faithfully represented in a 15 MP image. The image from that sensor may have other merits (such as pixel-level detail and lack of some artifacts) and it would be better for Sigma to talk about those rather than make this inaccurate claim.</p>

<p>Since it is a DX sized sensor but the camera is about as expensive as a 1D X or a D3X ... that just means that virtually no one is going to place the bet to check it out. So any benefit will be unknown.</p>

<p>Frank, photosites are not pixels. The SD1 produces 14 MP images, period. To argue that it produces 43 MP images is to claim that it has higher resolution than a Pentax 645D, which is just crazy.</p>

<p>As for "applying my reasoning the other way"... I already did (see above). Bayer demosaicing does reduce resolution somewhat, but if you think a 14 MP Bayer-sensor camera only has 4 MP of real resolution, you don't know what you're talking about.</p>

<p>Look at http://en.wikipedia.org/wiki/Foveon_X3_sensor for a discussion of this type of sensor.</p>

<p>It is not comparable to other sensors in the <em>counts</em> of "pixies" regardless of its benefits or debits.</p>

<p>Mickey, good question. Others are correct to point out that the resolution is really only 14 MP, but it's perhaps 14 MP with better color. Maybe. I would be very surprised if the Sigma digital cameras don't have antialiasing filters, though. AA filters are useful for any digital medium, whether for imaging, sound, or whatever. That doesn't mean a camera can't work without one, and in fact there are companies that will "hot rod" a camera by removing the AA filter. However, you don't want to take a picture of a screen window or a piece of fabric with such a camera!</p>

<p>That said, I offer these answers to your questions:</p>

<p>1) Maximum useful pixels is to some extent (and for various reasons) tied to format. We have already exceeded the number of pixels useful in compact digital cameras, and further MPs are simply sales hype. We are probably pushing the limits of maximum useful pixels in DSLR cameras, but that leads to your next question:</p>

<p>2) Some of today's best lenses are capable of outresolving the highest resolution sensors sometimes, under the right conditions. You have to be somewhat anal (can I use that word here?) to tap those resolving capabilities. Forget zooms. Forget hand holding. Forget shooting at any aperture except around f/8. You'd better use mirror lock-up. If you do all these things and have a sufficiently sturdy tripod, sandbagged, with cable release, the lens might outresolve the sensor in the middle of the frame. However, if you shoot like me, 12'ish MP is probably adequate, and 24 is just comfy. Bill Gates once supposedly said 640 KB was more RAM than anyone needed. In that same spirit, I think 24 MP is probably enough for most full frame DSLR needs, and anything further would just be sales hype. (That's just my opinion.)</p>

<p><em>Forget zooms. Forget hand holding. Forget shooting at any aperture except around f/8. You'd better use mirror lock-up. If you do all these things and have a sufficiently sturdy tripod, sandbagged, with cable release, the lens might outresolve the sensor in the middle of the frame.</em></p>

<p>It's not quite like that. You do have to use a good tripod (or flash) to get the most out of a high-resolution sensor ... but in reality you have to use it anyway, to get the most out of a lower-resolution sensor unless you are using very high shutter speeds. It's not like using a tripod is "anal"; it's normal practice e.g. in landscape and architectural photography; you have to use it to be able to combine multiple exposures to do stiching or to increase dynamic range in a precise way (so that you don't have to "fit" the images and then crop a lot of extra that doesn't fit the aim of making a rectangular final print). You have to use it because you need a lot of depth of field. You have to use it because otherwise applying tilt correctly would be a shot in the dark; you'd never get it right hand-held. And finally you have to use it because you want to get the maximum dynamic range in a single capture so you need to use base ISO for that. If you want controlled subject motion blur (just in the right places, e.g. water) you need a tripod. For architectural work you need the tripod also because you need your lines straight. I could go on and on ... a tripod is not some act of desperate perfection but the only way to do proper work in these fields of photography.</p>

<p>f/8 is not typically the sharpest aperture with high-resolution sensors and the better lenses. f/4 and f/5.6 are more typical at least with fast lenses. f/8 is still acceptable but it's not quite as crisp as f/5.6 - f/8 may have better FX corners than f/5.6 but not the centre.</p>

<p>If you look at photozone results of the D200 vs. D7000 with the same lens you can see that the 16 MP DX sensor of the latter easily outresolves the 10 MP sensor at all parts of the frame at all apertures even with common mid-grade lenses e.g. 70-300 VR. Yes, of course they used a tripod and focused carefully as any careful photographer would. Even if you can see the difference between center and edge with your current camera, both center and edge will still be improved by a higher resolution sensor (assuming proper technique and optimal aperture, and base ISO). I can confirm that the D7000 gives excellent results at ISO 100 on tripod and in many conditions at far less restrictive settings, but the limitation there (for me) was autofocus - in many cases I simply couldn't get the kind of consistent quality I get with my FX cameras with the D7000. But that has nothing to do with the limitations of the main sensor or lenses. The sensor is excellent and it is easy enough to find some lenses that make the most out of it.</p>

<p>Whether any of this matters much for your own type of photography is another matter of course. The low-pixel count sensors such as D3s have their own advantages which may be more important than higher resolution. For 95% of serious photographers a moderate pixel count is probably better than an ultra high one, if only for practical reasons of storage space, processing capacity, and if you shoot moving subjects without flash it can be more difficult to get benefits from the higher resolution sensor because of limitations in shutter speed, ISO and especially autofocus speed and accuracy. I have been quite happy with 12 MP FX for almost four years now - these cameras are great for general photography and especially people photography in uncontrolled light, but they're far from ideal for landscape, studio, and architectural applications. I think 18MP FX would be an ideal compromise for my needs and I'm hoping the D800 and D4 will be along these lines (as is the new Canon 1D X).</p>

<blockquote>

<p>I would be very surprised if the Sigma digital cameras don't have antialiasing filters, though. AA filters are useful for any digital medium, whether for imaging, sound, or whatever.</p>

</blockquote>

<p>The Sigma SD1 does not have an anti-aliasing filter. Neither does the Leica M9.</p>

<p>While it may be argued that an AA filter is "useful for any digital medium", you have to balance that against the cost (reduced sharpness, since an AA filter works by subtly blurring the image). The biggest reason to have an AA filter in most digital cameras is the Bayer matrix and the associated demosaic logic, which can introduce ugly moire effects when interpolating colors. Since the SD1 doesn't have a Bayer matrix, the need for an AA filter is greatly reduced. On the other hand, the Leica M9 also doesn't have an AA filter, and I haven't seen anyone complaining about all their shots being ruined by moire.</p>

<p>Ilkka, I don't disagree with you. What I claimed is of course controversial. I'm not saying "anal" in a bad way. I perhaps should have said "extreme attention to detail," as it is so easy to introduce the tiniest blurs that would render all of the superior resolution of one's lens and sensor moot. I'm also NOT saying there's no sense in using a tripod. If it says anything, I own and use a few of them, and some of what I do would be impossible without them. However, most of what I do is handheld, so for me the gazillion-pixel resolution capabilities don't compute anyway. </p>

<p>Like you, I find I'm fairly happy with 12 MP, but I'd like a bit more for those infrequent occasions where I do try to push the limits of what my gear can accomplish. My next upgrade will be from the 5D to the 5DII, but most of what draws me to that camera is the lower-light capabilities, which are very important to me.</p>

<p>If I had to guess, I'd imagine these megapixel wars will eventually crawl to a stop somewhere in the low 40's for full frame DSLRs, beyond which increased pixel density would just be senseless, due to realities of diffraction. I have faith that serious photographers buying serious DSLR cameras won't tolerate much more than that and will be pushing for better pixels, not more pixels. Nikon has already taken this direction, to their credit. </p>

<p>Looking into my crystal ball, I suspect APS-C cameras will be made to match the full frame cameras in pixel count. Then compact sensor cameras will eventually bring up the rear, struggling greatly to deliver the same pixel counts as a DSLR, pushing the limits of chip design and manufacture to stupid levels, for absolutely no benefit, and at some cost to image quality.</p>

<p>... all speculation, of course!</p>

<p>Craig, I read the following from dpreview about the M9's sensor:</p>

 

<blockquote>

<p>A moiré filter has not been integrated, allowing full exploitation of the superb resolution of Leica M lenses. Any moiré patterns occurring are eliminated in the camera’s signal processing software. The optimised signal–noise ratio reduces the need for digital post-processing, and results in high-contrast, high-resolution exposures with natural colour rendition from corner to corner.</p>

</blockquote>

<p>All I can say is that handling moire patterns in the signal processing software, and not prior to the sensor, is utter nonsense. However, who am I to argue with the likes of Leica? (Well, I've actually studied this sort of stuff quite a lot in graduate school, but I'm a former sensory physiologist and current photographer, not an engineer, so most people would feel I'm wholely unqualified to question Leica in this regard.)</p>

<p>What the M9 and the foveon sensor share is a large fill factor, due to their architecture. It has nothing to do with Bayer de-mosaicing. (Note here: The M9 sensor has a Bayer array.) A large fill factor helps to reduce moire; however, it doesn't eliminate the issue. What does eliminate the issue is sufficient oversampling. If the sensor's resolution exceeds the lens' resolution, then the lens cannot project spatial frequencies that would result in moire patterns in the first place. I can frankly see why moire would rarely be an issue in a 14 MP Sigma camera that can only take Sigma lenses. (Sorry, Sigma fans!) However, I'd be a bit surprised if that's the case with an 18MP full frame Leica. Maybe the Leica employs moire detection that is somewhat like face detection in other cameras. If so, that's a pretty magical trick. However, I'd be surprised. Something's gotta' give.</p>

<p>Sarah, I doubt the people at dpreview know enough about either sensor technology or journalism to be reliable sources of information about this sort of question. At best, they're probably indirectly quoting some Leica marketing rep who doesn't necessarily understand the issue either. I was just citing the M9 as an example of a camera with a Bayer matrix that doesn't have, and seems not to need, an AA filter.</p>

<p>If the sensor outresolves the lens, then in effect the lens is its own AA filter; that is, if the image from the lens isn't sharp enough to outresolve the sensor, then obviously you won't have moire because the image is blurry to begin with. I doubt that's Leica's solution, though, considering how scary-sharp many Leica lenses are. Nor do I really think there is software to detect and remove moire built into the M9. More likely, Leica just found in testing their sensor that moire crops up very infrequently, so they decided to just leave out the AA filter.</p>

<p>Probably right.  However, when moire does crop up, it can really ruin your day!</p>

<p>Thank you all for addressing and expanding on my post. Your responses are exactly the reason I decided to become a member of PN! There is so much expertise out there that its comforting to know so many experts chose this site to share that expertise and talent. Thanks again. Warm regards for the Holiday Season! </p>

<p>><br>

However, you don't want to take a picture of a screen window or a piece of fabric with such a camera!<br>

><br>

Ever worked with a Leica M9 Sarah? It uses no AA filter. Moire patterns are easy to eliminate when they crop up.</p>

<p>><br>

I think 24 MP is probably enough for most full frame DSLR needs<br>

></p>

<p>Speak for yourself, some of us like to print large without interpolating. I keep reading these kind of statements over and over and I think they are baseless.</p>

<p>Sorry, I'm not trying to be argumentative just positing a differnt point of view.</p>

 

<p>Don, no, I've never shot with an M9. I'm a Canon photographer. I own and occasionally use a Leica IIIf, if that softens your contempt and somehow makes my commentary worthy.</p>

<p>Now if you go back and search some threads in the digital darkroom forum, you can find very long discussions where photographers very skilled at postprocessing went through very heroic efforts to save the photos of some of our professional colleagues when moire patterns cropped up. I especially remember a difficult color/luminance moire pattern that cropped up in the fabric of a suit worn by the VIP at one event that could not be re-shot. I really wish you had shown up to contribute your postprocessing skills to that discussion, because it had quite a few of us scratching our heads for over a week. (I eventually solved the problem, somewhat less than satisfactorily, but it wasn't easy.)</p>

<p>I have no problem with someone else desiring more than 24 MP or expressing an alternative point of view. I do, however, have a problem with people telling me my opinions are baseless, when in fact they are not. I'm not feeling too warm and fuzzy about the tone of your post either.</p>

<p>><br>

Don, no, I've never shot with an M9. I'm a Canon photographer. I own and occasionally use a Leica IIIf, if that softens your contempt and somehow makes my commentary worthy.<br>

><br>

You missed my point (my fault actually.) The M9 has no Bayer filter and moire does occur but can be corrected with PS actions when it happens. I didn't invent them but others have tackled the problem successfully. There was no contempt (intended) in my post. </p>

<p>I've noticed you are very thin skinned Sarah, when others disagree with you. I usually don't post warm and fuzzy but just get right to the point, though I'm not trying to tick you off.</p>

<p>Anyway, sorry I bothered expressing a contrary opinion.</p>

 

<p>The problem occurs when there is detail that must be shown in the area of moire, such as the curves/wrinkles/folds of the fabric. It's not the same as blurring out detail in a grill (e.g. on an air conditioner) where no detail is needed. I considered "hot-rodding" my 5D (having the AA filter removed) when it suffered an AI/IR filter scratch. I opted against it simply because of the potential moire issue. If all I did were nature photography, I don't think I'd hesitate. There are fewer regular patterns in nature, although they do sometimes exist.</p>

<p>There is a related issue not discussed here. There are two types of moire -- luminance and color. Our AA filters are generally effective for the former and not the latter. That's where the folks with AA-filtered sensors get their occasional headaches. I honestly can't imagine dealing with luminance moire too. Leica has some very sharp lenses, and I suspect the moire issue crops up with the M9 a lot more frequently than they care to admit. In some cases one can do a re-shoot, which is professionally embarrassing. In others a reshoot is impossible, and a pro photographer can only apologize to the client. That's never a good situation to be in. That's why many of us went to such great lengths to help out another photographer when a particular pattern in the client's suit turned psychodelic (the aforementioned thread)!</p>

<p>PS Yes, I'm a bit thin-skinned. Please accept my apologies.</p>

<p>Craig, I heard that the Sigma SD14 was not 14 megapixels too. WRONG! It IS, and you can shoot 14 megapixel JPEG photos with it, by setting the image quality to super fine mode. They are very good quality images. You can see one at the beginning of my analysis gallery:<br>

-<br>

http://ffphotos.zenfolio.com/SigmaSD14<br>

-<br>

My estimate is that the Foveon sensors actually do perform well enough to claim the resolution that first Foveon and now Sigma claim. In red scenes, you can see that they perform incredibly well, compared to Bayer pattern sensors (standard sensors). My EXPERIENCE with my SD14 has taught me that using the "interpolated" super fine mode is not as good as just using the RAW mode and creating 4.65 megapixel JPEG images from RAW files, so to a degree, you are correct about the resolution of the BEST images the camera is capable of producing (and I am sure the SD1 is the same). Ultimately, I think other factors, such as noise are going to be as important as resolution. At ISO 100 (the setting where I shoot the majority of my photos), the Sigma SD1 seems to produce the least noise of any APS-C sensor camera, and that includes the new A77 from Sony.<br>

-<br>

Here are some images and an article, for anyone reading here:<br>

-<br>

Notice that the following two images make an attempt at showing the ultimate resolution possible from the sensor. The middle of the images shows the best resolution of the lens used on the camera better than the edges. IF you look at the edges of the images, you will see the limitations of the lenses. The zoom lens on the Sigma SD1 shows lots of blur and CA in the corners of the image, while the prime lens on the Nikon D3x does not. Of course, you should take into account the fact that the center of the image could be affected by the lens too, so the Nikon D3x might actually be capable of performing better, if a better lens were used. (I doubt it though, since pixel density is less than many of the other cameras that have been tested with the same lens. The lens limits do not seem to have been reached yet.)<br>

<a rel="nofollow" href="http://216.18.212.226/PRODS/D3X/FULLRES/D3XhRES6048F.JPG" target="_blank">http://216.18.212.226/PRODS/D3X/FULLRES/D3XhRES6048F.JPG</a><br /> <a rel="nofollow" href="http://216.18.212.226/PRODS/SD1/FULLRES/SD1hRES4704F.JPG" target="_blank">http://216.18.212.226/PRODS/SD1/FULLRES/SD1hRES4704F.JPG</a><br>

http://www.dpreview.com/galleries/5418833960/photos/1390445/sd1-a77-resolution<br>

<a rel="nofollow" href="http://www.rytterfalk.com/2011/01/20/why-i-choose-sigma" target="_blank">http://www.rytterfalk.com/2011/01/20/why-i-choose-sigma</a><br>

-<br>

Good luck with you research Mickey. Just so you know, I plan to get a Sony A55 (at high ISO there seems to be no significant difference between image quality from the A55 and the A65/A77), and then I will save up for a Sigma SD1. At least that's the plan. I bought the SD14 to test the theory and find out for myself how the Foveon sensor performs. Now I am a convert, like when I switched to Apple a few years back. There is nothing like trying it out for yourself.</p>