Read fascinating and revealing interview with head honcho at Canon Cameras, Masaya Maeda:

http://www.dpreview.com/news/0810/08100302...oninterview.asp


Best,
Bruce

That seems to explain some of the reasoning behind the 5D2 using what many consider to be an inferior AF system. I don't know if it will sarify the naysayers.

Perhaps you could get this discussion started by specifying what you think is "fascinating and revealing" about the interview.

I think the thing that was most interesting was the discussion about reaching the limits of diffraction.


There was a lot of discussion here about when pixel-pitches get so small as to outresolve the lens. At that point you are only capturing fuzzy detail, so why bother. Canon didn't seem to offer any spin on this fact and later admitted that they may need to improve their lens lineup to deal with this.

At first, digital cameras didn't get as much from the lenses as film. Now it is getting to the point where the digital cameras are easily outresolving the lenses and may demand upgrades...what a great new revenue stream for Canon. I can see it now "Canon 50mm f/1.0 L-Pro."



I also was disappointed about their commitment to anti-alias filters. I have seriously considered sending my camera off to have the filter removed. If I get the 50D, it might be tempting to have my 30D filter removed. Too bad it is soo expensive.

What fike said.

"We believe the users benefit from being able to trim or crop their pictures, and also to produce large prints, and most of these benefits we aim to provide by increasing megapixels."

I believe this is a mistake driven by simplistic marketing. This was a reply to a question about the G10 and its excessive number of megapixels, but it applies to the 50D and 5DII as well. Consider that Maeda san is arguing having extra megapixels is primarily a solution to poor framing or lack of lens reach. Well greater detail is gained in larger prints if the lens is kept at or near wide open, it comes at a cost of noise. Now you pay the price of larger file sizes for every image you take, even when you will only need those extra pixels very infrequently to print really large. Ironically, Maeda san concedes that diffraction at smaller apertures also limits the efficacy of these extra pixels and when pressed on this he said, "To some extent I agree with you, which is why we're looking at the possibility of adding diversity to the G10, which would be the answer to those looking for something other than high megapixel count." However, there is no indication whatsoever, anywhere in Canon's line that they will go in this direction.

Tony,

What you, DPR, and others who naysay larger pixel density seem to disregard in your arguments is that sensor technology leading to lower noise, greater sensitivity, and deeper wells is evolving and not static. We have been through this discussion many times, notably most recently on the long 5D Mark II announcement thread. It is not as though the science is currently up against any fundamental limitations of physics, as far as I know. The quality of the 5D Mark II pixels is not to be confused with either the quality of 5D pixels or even of 1Ds Mark III pixels, for that matter. Maeda-san also makes that point in the interview. (That was one of the "revealing" aspects of the interview, by the way, in partial answer to your earlier question... in essence an admission that the 5D Mark II surpasses the 1Ds Mark III IQ.) And, Canon technical press releases provide a not insignificant amount of detail as to how Canon has achieved higher pixel performance while increasing density.

And yet, one after another of your forum statements appear to presume that pixel performance (per pixel area, let's say) has been and continues to be static. That is not a reasonable position. In as much as you seem to be Nikon-centric, think of it this way: If IQ per pixel area were static, why is it that D3 IQ is generally considered to be superior to 5D IQ? Reason: A two-year plus jump in sensor technology.

I fear that you are unreasonably painting yourself into a corner. What are you going to say when the reviews are published showing the 5D Mark II IQ to be equal or superior to that of the old, 12 mp 5D? Maybe you will simply disagree with the judgment of the reviewers? Judging by the queues that are forming to buy the 5D Mark II, it appears that most people get this.

I predict that within 10 years the MF market will have all but dried up because the 1Ds Mark xx (and Nikon equivalents, of course) will be at current-day MF IQ at 40mp and very few photographers will care to spend lavish amounts of money for the tiny incremental gains that MF will provide.

Bruce

As far as I'm aware, Canon have never produced a DSLR which has more noise than a previous model with fewer pixels, when equal size images are compared from same format DSLRs.

Comparing the noise of individual pixels is another matter. The larger pixel then usually has the advantage, but not always. I recall the dpreview comparison between the 1Ds and the D60 which demonstrated that a 1Ds pixel was actually slightly noisier than a D60 pixel, up to ISO 400. However, because a 1Ds image consists of substantially more pixels than a D60 image, noise in the 1Ds image is less noticeable when comparing equal size prints or equal size screen images.

When Canon reaches the point where increasing the number of pixels would actually increases the amount of visible noise in the image as a whole, then that's the point where they should stop.

I don't believe there will be any circumstances where a 15mp 50D will show more noise or fuzzier results than a 20D, 30D or 40D. However, there will be circumstances where a 50D image, using a good quality prime at its optimum aperture, will produce slightly sharper and more detailed results than the previous cropped format models mentioned. (It is assumed in this statement, that we are referring to the same subject and the same lens from the same distance to subject, used at the same aperture and the same shutter speed.)

You can add the head of the entire camera division at Canon to the list, Maeda san pretty much conceded that higher pixel density limits the resolution attainable at smaller apertures. This isn't even up for discussion among anyone who understands the most basic fundamentals of the optical properties of lenses and apertures.



Right, as far as you know, which is because you have not yet grasped the fundamental limitations of diffraction and how it is recorded by more tightly packed sensors.

Here's the point about the 5DII being better than the 5d in terms of noise -- I never said it wasn't; however, we'll find out eventually when the camera is actually in the hands of its many enthusiasts. My point is that the 5DII could be better than it is if it had larger photosites, and that's a trade-off that Canon has made to pander to those that seem to think that the solution to their photographic needs is to have more pixels. Yes, at wider apertures you will be able to use those extra pixels if you print large enough, but 99% or more of all prints coming from these cameras will not require anywhere near the number or pixels they generate. For most users the price of the one or two huge prints they adorn their homes with will be doubling the size of every file they save on a memory card, store on their hard drive, edit in Photoshop, and have to archive.

Now what I found interesting about the interview was that Maeda san said Canon is considering delivering lower MP cameras to address the needs of photographers who are not caught up in the "more is always better" mentality that grips the industry at this time. Peculiarly, you wrote that you agreed with Marc, but then you turned around and contradicted that by taking issue with me. Perhaps you are letting your previously formed opinion about me cloud your judgment; anyway, I will spare myself from arguing with you any further on this and add you to my "Ignore User" list, which I am finding to be quite liberating.

Optical low pass filters are essential for sampling theory, the basis of digitizing the image coming into the camera to work. They're doubly necessary on a bayer pattern colour filter array camera to avoid the worst of chroma moire and other issues. To ask Canon to not put in an OLPF is to ask them to incorrectly engineer their camera, and although we all have complaints about features of Canon cameras (useless print buttons and mirror lockup lacking a specific button) Canon do engineer things correctly.

If you don't want an OLFP buy a Sigma or a Leica, or one of those medium format cameras that show moire everywhere.

Graeme

I don't have one on my D200 converted to IR, and I have yet to see any moire on it.

Oops; sorry Tony. I did not mean to offend you.

Best,
Bruce

I'm not offended, I just don't have time to read posts like yours. Being flippant does not compensate for attributing things to me that I did not write; or for exhibiting a lack of understanding of what the limitations are to increasing pixel pitch density, or the costs of generating larger files that will prove unnecessary for the vast majority of users the vast majority of the time.

Same here with an XT.

Canon's release of the 5DII with the 21MP Full-frame sensor, was not "pandering" to anyone. The one and only reason why Canon released the 5DII with the 21MP was due to Sony twisting their arm by releasing the A900 Full-frame, with the 24.6MP sensor (announced almost a year back).

Simply introducing a 21MP sensor is just half the story. This 21MP will need high-end processors to crunch the massive amounts of image data that will come off the sensor - hence taking a hit on FPS. Also, once the data is processed, the data pipeline further downstream, will need to be robust enough to transport such gargantuan data volumes (especially when shot at say 5FPS at 21MP resolution etc) from the capture area to the buffer/CF-storage.

All of this high-end componentry that makes all of the above happen costs money, and Canon would have frankly preferred to have come out with a 16MP sensor than the more challenging 21MP sensor - believe it or not !

When Sony's Full-frame camera comes with a 100% viewfinder that is larger than the Viewfinder of the 1DSMKII and the Nikon D3, with IS built into the body thus stabilizing the 35mm primes, the 50mm primes, the 85mm primes, the 135mm primes, the 24-70 f/2.8, the 16-35 f/2.8 etc., along with 5FPS at 24.6MP resolution and pairable with ultra-high-end Carl Zeiss lenses (with Auto-focus), Canon too have to step up their game, since the "full-frame body marketshare" has a strategic impact to long-term viability of the system. When an electronics giant like Sony snaps up a traditional Photography company like Konica Minolta and then introduce a Full-frame a couple of years later, Canon knows that they don't intend to play second fiddle for too long and they have started to push hard......."pandering" was the last thing on Canon's mind, when they were forced to introduce the 21MP FF sensor in the 5DII, to be quite honest.

Why ? refining a already designed 21MP sensor is a LOT cheaper than designing a new 16Mp sensor.

OTOH I have seen moire in photos from an unmodified D200.

Your attitude towards diffraction seems to involve a bit of misinterpretation. Any given pixel density has a certain diffraction limit, yes. On, say, the 6mp rebel, lets say you could shoot at f/16 and not get diffraction limitation until f22. Then lets say the 10mp xTi can only go to f11 before you start seeing the resolution drop. And lets say (this probably isn't accurate, but we're being hypothetical) that with the 15mp 50D it shows up at f8. The thing is, at f11, you'd still get the same resolution as the xTi would, and at f16 you still get the same resolution as the rebel. In this scenario, the most resolution you can get from APS-C at f16 is 6mp equivalent, the most you can get at f11 is 10mp equivalent, and the most you can get at f8 is 15mp equivalent. Sure, you could say "we're only going to make it 6mp so that we don't see diffraction until f22" but then you're also limited to 6mp at all the larger apertures. It's not as if diffraction is an effect of the pixel density, it's a property of the light coming through the lens. It's just that we can see the effects more clearly when we have greater pixel densities. If it were, you would get more resolution from the rebel at f16 than you would from the 50D. But you won't you'll get the same.

So why would manufacturers limit pixel density for fear of diffraction? Higher pixel densities will achieve the same resolution at smaller apertures as the lower pixel densities would, and they'll achieve higher resolution at the larger apertures before diffraction comes into effect. Perhaps someday manufacturers will implement a feature that downsamples to account for diffraction depending on the shooting aperture. In this hypothetical, the 50D might downsample all RAW or jpeg files shot at f16 to 6mp, and all shot at f11 to 10mp, like sRAW. That way you wouldn't have to deal with storing and moving around the extra data that isn't really helping you.

I would take exception with the first sentence of your reply as my interpretation is essentially the same as the rest of your reply. As to why would manufacturers limit pixel density, well to keep files from becoming unnecessarily large.

In my opinion, Canon's three most recently released cameras are all suffering from excessive megapixel syndrome. I think The G10 should be going in the other direction completely, 6 million quality pixels would be much better for such a small sensor. I think the 50D would be good for use with longer focal lengths when greater reach is desired, for most portraiture though it would be overkill, and for landscapes it would be no better than the 40D for me because I simply have to stop lenses down to around f/11 or the image circle is too small and the lenses' edge performance is challenged. For the 5DII, I would want a cropped mode as not every image needs to be full frame or 21 MP.

My tests have shown so far that the higher pixel density of the 50D actually shows slightly more detail than the 5D, even at F22, using the Canon 100-400 at 400mm from the same position. Of course, the increase in detail is more obvious at F8, thus confirming that the real benefits of the high pixel-density cropped format are as a lens extender with no exposure penalty.

However, I should mention that such increased detail at F22 would be noticeable only in a very large print upon close inspection, or at 100% enlargement on the monitor. It's more clearly noticeable at 200% enlargement and at 300%, hits you in the eye.

Tony seems to be under the false impression that there's a brick wall regarding diffraction. At one F stop there's a marginal increase in detail, then one stop further down there's zilch. That's not so.

Tony, if the large files are the problem for you, why not just use sRAW? You're probably going to get a better file from a 21mp camera downsampling to 10mp than from a camera that shoots 10mp natively due to the benefits of oversampling.

Ray, that's really interesting. The possibility that cameras with higher pixel densities can fish more resolution out of the sensor than a lower-density sensor even when they're both under their diffraction limits is a strong argument for increasing pixel counts as long as noise and DR are not negatively impacted. If we're still using bayer sensors in 10 years, I wouldn't be surprised if our cameras have 40+mp sensors that clearly outresolve our lenses (allowing us to get rid of AA filters and let the lens' sharpness shine through) and most photographers set theirs to downsample to 6-12 mp for most applications. The benefits of oversampling are well worth it if manufacturers can bring down the costs and get DR under control.

I think it's quite likeley we shall eventually see a 40mp full frame sensor. The cropped formats have always led the way regarding pixel density, and full frame sensors have caught up a few years later. The 5D MkII and 1Ds3 have the pixel density of the 8mp 20D.

DR and noise seems to be largely affected by sensor size. First indications are, the 50D has no 'real' DoF advantage compared with the 5D. In other words, F4, 28mm and ISO 200 with the 50D is no better than F8, 45mm and ISO 800 with the 5D. Both sets of images seem to have about equal resolution and equally low noise.

There might be a slight advantage at very high ISOs, such as the 50D at ISO 3200 compared with the 5D at ISO 1600 underexposed 2, 2 1/2 or 3 stops. The DoF equivalence does not seem to correspond exactly with the 1.6x crop factor. It tends to range between 1 stop and 2 stops' difference depending on distance to subject (and possibly lens design).

I doubt Canon has thrown away the design of the 1DSMKII's 16MP Full-frame sensor, which was cutting edge less than 2 years back. They just had to brush it up a bit and with some additional tweaks using the latest technology, it could have done its stuff within the 5DII.

The advantage would have been that using the same electronics, they could have easily achieved over 5FPS (assuming the shutter is upgraded for 5FPS), since the data coming off the sensor, would have been significantly less, when compared to a 21MP sensor.

Either way, the deed is done and I doubt Canon would go back to employing a lower pixel density FF sensor in future products. I personally would have preferred a 14-16MP Full-frame sensor than a 21MP or 25MP Full frame sensor - but that is just me.

It isn't just large files, it's also unnecessary format space. Often we want to just use the center of the image for telephoto lens shots, and it precisely when doing that sort of shooting that added fps would come in handy. That's why I prefer Nikon's cropped format approach, which I expect to see in their high MP FX DSLR.



I strongly suspect that many of us won't be using BFA DSLRs in even 5 years: http://www.imaging-resource.com/NEWS/1186694099.html Three color photosites are coming, and they will be coming from several directions as I'm sure Canon also has plans to do this. Imagine 12-16 MP 3 color DSLRs that effectively capture approximately double the resolution of the current BFA DSLRs; the advantages will be higher native ISO, more accurate colors, no more diffraction limitations, smaller files (in a time when computing power will be even greater and storage even cheaper), and the only limitation on fps will be if the AF can keep up (that will also increase as the camera's processing power increases).

Yes. And this is why these discussions about mp vs. noise vs. image quality always go around in circles. It's funny to me that when we discuss how to get the highest "technical quality" photographs out of our equipment, we all agree that your entire workflow / technique matters. From accurate focusing, tripods, MLU, correct ISO for the subject, DOF vs. diffraction, camera selection, lens selection, ETTR metering, blah blah. All these things add up and contribute to an image's technical success.

Yet when a new camera comes out, we tend to grab on to some specific feature/improvement and evaluate it to death, treating it as if every other input to the IQ process is either static or zero (ex: Nikon's high ISO marketing, Canon's high MP marketing). Often before there's any way to confirm, evaluate or prioritize the improvement.

Not only that, but the priority of these inputs is subject-dependent, as Michael pointed out in his equipment selection for Botswana. So people value different features/improvements that affect IQ differently. From polling thousands of posts, I have confirmed that Canon shooters always print > 40"x60", and Nikon shooters always use ISO > 3200. :~) [if I may borrow Schewe's logo just once]

Would the 5D be "better" with 16mp instead of 21? We'll never know because they didn't make one. My guess is in most cases it would be awfully difficult to tell anyway. But I'm sure someone will devise a test that proves yes. And no.

Dave Chew

Move to an approach like you post to won't help matters significantly. The primary reason a Sigma with it's foveon has a higher measured resolution is that it omits an optical low pass filter from it's design, and hence allows through all manner of nasty moire and aliasing artifacts. Hence a lot of it's resolution is fake, and a lot of it corrupts real lower frequency data. As pointed out above Canon fully understand the necessity for correct optical low pass filtering and will not make a camera that aliases nastily. Also, such a method is still using the same silicon area for the three receptors as you'd use in a CFA design - it's essentially an RGB stripe pattern CFA with a complex lens structure on top of it, so to get higher resolution, you'll have to move to smaller photosites and higher noise. Assuming you have the technology to make those smaller photosites less noisy - at that point, placing them in a Bayer CFA will give you more perceptually relevant resolution than the stripe+dichroic mirror approach. Indeed, unless you really like the way a dichroic mirror splits up the colour spectrum, you may as well just go with a plain RGB stripe CFA and let the optical low pass filter ensure that each triplet gets pretty much the same light hitting it.... And have a lot cheaper sensor with similar noise and resolution characteristics.

I take Canon interviews with a piece of salt as past performance has shown them to be more of a marketing exercise than anything else.

As for high MP, it is also more of a marketing wheeze than anything else. They know that consumers compare specs, and the most obvious one is the MP count. That is why Point and Shoot cameras now have absurd pixel counts. In practice a user will not be able to realise the MP count of the 50D unless using a small number of high performance lenses. Others will have so much CA and softness in the corners as to make the exercise pointless. Most example pictures on sites such as DPREVIEW are awful. Even with the best lenses, you will only be able to use certain apertures. Stop down to F11 for DOF, and the chances are the limitation will be the lens, not the sensor. For macro work you generally need to stop down to F11 at least, in which case there is not much point going beyond 12MP. Oh, and to realise the resolution you will need to use a tripod of course, or high shutter speeds.

Canon should IMO have stuck with 12 MP, or even 10MP, and concentrated on improved noise and DR. They would then have had a class leading camera with real advantages. As it is, they simply have more MP which most of the time is of no use, and yet file sizes are bigger. Oh, and the need to shunt more data through the circuitry means you need more complex electronics, hence putting up the price. Smaller file sizes might have allowed faster frame rates too.

Not so long ago people were responding to the Nikon D200 by saying that the 8MP of the Canon 30D was more than enough, and 10MP was too much. Hi ho.

Dave,
The 50D can be considered as a crop of a 40mp full frame 35mm DSLR. Even with a medium quality zoom like the Canon 100-400 IS, 40mp is more detailed than the 12.7mp of the 5D, not only at F8 and F11 (the apertures at which this lens is sharpest) but even at F22, although the improved detail at F22 is very slight.

Below is the full scene followed by 200% crops at the plane of focus, at F22 and F11. At F11, the 5D is not even capable of showing the corrugations of the roof.

These image have had only default conversion in ACR plus a slight adjustment of brightness to equalize the appearance. ISO was 400 for both cameras.


Click to view attachment Click to view attachment Click to view attachment

Click to view attachment

What you show is hardly surprising since the 5D corresponds to ~5MP when cropped, though the difference between the images with the lens at F22 is as good as nothing IMO. I thought the issue in this thread was whether or not 15 MP was an advantage compared to ~12MP.

I can think of two things that are significantly improved.

First is file sizes will be smaller; you can uprezz in software and use that function to eliminate aliasing.

Second is diffraction limitations can be moved a stop or more higher, which delivers more practical resolution at higher f/stops.

Uprezzing in software doesn't eliminate aliasing.

Diffraction limiting is down to the size of the pixel. If the pixel gets smaller, you've just traded resolution for noise.

If you have the same number of photosites in a different arrangement, then you get the same file size (assuming no compression).

Graeme

Whether one is comparing 15mp with 5mp at a cropped-format size, or 39mp with 12.7mp at a full frame size, the differences are the same because the pixel densitiies are the same.

Whether or not you think the extra detail at F22 is worth anything, is a matter of opinion and will depend on your purposes. I'm merely presenting the facts and demonstrating that there is no brick wall effect within the aperture range that most 35mm lenses have, as one stops down, even with a 15mp cropped format sensor, and by extrapolation, a 39mp full frame sensor.

As a matter of fact, I can think of circumstances where that extra detail at F22, most apparent in the vertical stripes of the curtains and the brickwork of the window sill, could make the difference of being able to identify a face, the make of vehicle or any object in the scene for espionage purposes or even general historical purposes, if that were the only shot available.

Edit: The other issue I haven't addressed is the additional detail at F22 which should be apparent at a sufficient distance away from the plane of focus. The purpose of stopping down is usually to gain a greater DoF, albeit sometimes at the expense of some loss of detail at the plane of focus. The scene I shot for this comparison was focussed at or very close to infinity. There's nothing in the scene which is outside of the hyperfocal distance range, with either camera.

I was a bit too quick with that comment. Only from F16 to F32 are the 50D images within the hyperfocal range at all points. From F5.6 to F16 there's a very noticeable progression of increased clarity in the parts of the image closest to the camera. This increase in detail even at F22 (compared with F8), at the closest point to the camera, is significantly greater than the increase in detail at the plane of focus in the F8 shot, compared with the F22 shot, at the plane of focus.

In other words, even with a 39mp full frame DSLR, there could still be a good reason for using F16 and even F22 when extensive DoF is a priority. The loss of resolution at the plane of focus, due to the effects of diffraction, appears to be less than the increase in resolution away from the plane of focus.

The first set of crops below shows the progressive loss of resolution at the plane of focus, moving from F8 to F32. The second set of crops shows the progressive increase in resolution at the closest point to the camera.

Click to view attachment Click to view attachment

Of course, it hardly needs mentioning, if the 400mm lens had been a high quality prime instead of a medium quality zoom, the crop at F8 (at the plane of focus) would have been noticeably sharper than the crop at F11, instead of being roughly equally sharp. And the crop at F5.6 would have been clearly the sharpest, instead of being marginally less sharp than the F8 crop.

The Canon 100-400 IS is a lens that definitely needs upgrading. For all I know, perhaps recent copies of this lens actually have been improved. There's no reason why Canon should announce each marginal improvement in their production processes and quality control. This design has been around for a long time.

Unfortunately you did not clearly label your images so we do not know what you are comparing. Could you explain what each photo was taken with? That would avoid confusion. Then I can comment.

I'm comparing resolution at various F stops with the 5D and 50D. The name of the camera, the f stop and the focal length appear at the top of each photo, except for the two groups of 6 images in my previous post which were all taken with the 50D to demonstrate DoF issues, that resolution even at F16 can be higher than at F8, at least somewhere in the scene.

There seems to be some fallacy that there is no point in using a 15mp cropped format sensor stopped down below F8 because there is no resolution advantage with current lenses. My tests indicate this is just not true. Other tests I've done recently, comparing the 40D with the 50D, indicate that a 50D at F13 can provide the same detail as a 40D at F8 at the plane of focus, with same lens, but significantly more detail away from the plane of focus, as a result of the greater DoF that F13 affords.