Limits to sensor resolution?

<p>The November issue of B+W Photography (London) stated that "Canon has developed an APS-H size CMOS image sensor that delivers an image resolution of about 120 MP. The new CMOS sensor has a pixel count nearly 7.5 times larger than that used in Canon's EOS iDs Mark III and EOS 5D Mark II. It also offers 2.4-fold improvement in resolution"</p>

<p>Wow! Those 120 MP must be cramped even closer than Japanese subway passengers at rush hour. Are there any limits to sensor resolution? What negative effect might this have on noise or dynamic range?</p>

<p>The chill factor is this: Why should I even think about upgrading my 3 or 4 year old M8 (10 MP), which gives quite nice images, to an expensive 18 MP current body (or similar upgrade thoughts for other cameras) when such quantum leaps in performance as the Canon news may be sooner than expected?</p>

<p>Because they don't have a processor small and fast enough to handle that data load. And also there is no commercial small power source/battery available that can provide power for it for any extended time. "Megapixels" have never been the limiting factor in digital camera design...</p>

<p>Why move to larger MP count? Very Good Question. However, searching for the larger sensor size makes more sense to me due to its affect on the focal point of the lenses. I too am questioning the value of such a large MP sensor as would be the new CMOS? We are obviously limited in size for what we are going to send out over the internet as well as toning down our file size for printing services that request only, (not more than), 300 dpi. I recently purchase the 5D ii and am thrilled to have it. But I still view so many images that seem superior to mine that are taken with "bigger stuff" (such as the new Hasselblad). I am new to this and have so much to learn. Am looking forward to the pros responses to these questions.

 

</p>

<p>The limit to sensor resolution related to the wavelength of light and the diffraction limited spot size.</p>

<p>These day 0.1 micron lithography is pretty easy and for a sensor that would be way overkill.</p>

<p>Of course as sensor size goes down, noise goes up.</p>

<p>Just using current technology, typical digicams have pixel density of around 50MP/sq.cm. That translates to an APS sensor with 165MP or a full frame sensor with 432MP - but with the noise performance of a P&S digicam.</p>

<p>Of course you only get a resolution increase if your lens is good enough and you only see that resolution increase if your print is big enough. So unless you are shooting in good light with $2000 lenses at their optimum aperture and printing 4ft x 6ft prints, you really don't need (and in fact can't use) ultra high resolution.</p>

<p>Canon's "plan" I think is to make a camera with ultra high resolution and then just crop the image rather than use different lenses. If you have a good enough lens, good enough noise performance and fast enough electronics with lots of fast memory, it's an achievable goal. In fact I think Canon have demonstrated prototypes (or at least mock-ups) of such a system</p>

<p>Valuable points! The Canon plan described by Bob makes sense, as they can use a high quality fixed focal length lens rather than a zoom, and process only "bites" (no pun intended) out of the overall image, which also have the benefit of needing less pixel processing. I like making prints of 16 x 22 inches, and larger, so I appreciate Steve's comment about the high MP Hasselblad large sensor providing better quality than a 20 or 22 MP full frame (35mm frame) sensor. What is needed in a 35mm full frame sensor to get, or more closely approach, that digital Hassy quality? Why (considering Bueh's comment) do we not already have it then? The battery energy and processing speed drawbacks?</p>

<p>3CCD might be an interesting dSLR approach if there was a way to deal with the resulting camera bulk.</p>

<p>I have an assortment of machine vision video cameras that deals with the same sort of problems with noise, cost, image quality and size, but big pixel size always wins in monochrome cameras while 3 (big) CCDs always wins in color cameras.</p>

<p>I don't see any reason why some hybrid technology can't be implemented to create a big 3CCD still camera, something like my Sony DXC-990 scaled way up. It won't be a suitable solution for every camera problem but it should offer certain advantages in many important areas. </p>

<blockquote>

<p>"Megapixels" have never been the limiting factor in digital camera design...</p>

</blockquote>

<p>Yep, I agree. The limiting factor from my own observations has been having a lens matched to a sensor at every focal length that allows it to deliver separation of color detail to each pixel. IOW when you capture a blade of grass at a certain distance the differences in hues of green as the sun hits it doesn't get rendered as one or two hues.</p>

<p>Of course the closer you get to recording just one blade of grass as in a macro shot will show much more differentiation in hues, so this does imply in a way that resolution plays an influential role in providing more detail. I would think a much more improved lens would have to be created to take advantage of this huge increase in resolution.</p>

<blockquote>

<p>high MP Hasselblad large sensor providing better quality than a 20 or 22 MP full frame (35mm frame) sensor.</p>

</blockquote>

<p>It's like the old quote about cars. "There's no substitute for cubic inches". In photography when it comes to pixels, there's no substitute for a large area.</p>

<p>The bigger the pixel, the better the signal to noise ratio. If you want big pixels and you want a lot of them, you need a larger format.</p>

<p>Yes, you can make very very good pixels and put them in a 35mm frame, but if you made the pixels bigger and put them in a larger frame, they'd be even better.</p>

<p>There's also the "good enough" situation. There comes a point when spending time and money of making something so much better than is needed (or is useful)that it's not really worth doing except as a technical exercise. See the Buggati Veyron for an automotive example. You really don't need 1000+ hp and a top speed of 265+ mph in a road car. Nor do most people have the $1.5 -$ 2 million needed to buy one. There's no earthly reason to make it except as an engineering exercise. It's brilliant, but essentially useless to 99.99% of the market. It's not even like there's anything in it that will "trickle down" to the family truckster.</p>

<p>You wouldn't want a 1GB 35mm DSLR (which is technically possible with current technology). There's nothing you could do with it you couldn't do with a 100MP DSLR or even a 25MP DSLR and you'd have a hard time storing (and processing) the data anyway. It would be impressive though!</p>

 

<p>Yep but someone Bob, will want to print 1 GP at 300ppi to make a very large detailed print and many others will convince themselves that they just can't do without 1GP even if they just do web uploads 1000 px wide. Thats just the way the pixel race works. Remember when many said 6mp was good enough and better than 35mm film....</p>

<blockquote>

<p>Are there any limits to sensor resolution? What negative effect might this have on noise or dynamic range?</p>

</blockquote>

<p>About 200 mp for an APS or 4/3 sized sensor, maybe 800 mp for an FF sensor. That gets us to the point where the sensor needs no AA filter, and the lens (even the best lens) provides all the AA "blur" we need, through diffraction and aberrations. It's a "diminishing returns" situation, the 800mp FF doesn't provide 10x the resolution of an 8mp FF, it simply gets us to the point where the lens is the only limiting factor in the electro-optical system.</p>

<p>It also leads to an improvement in color accuracy. As we approach that lens limit point, we don't need Bryce Bayer's brilliant idea to have twice as many photopic green "luminance" pixels as red or blue chrominance pixels any more, so we go to equal numbers of four different colors, and the color accuracy increases, sort of like Fuji Reala. Sony actually tried this a few years ago on the 828, but their camera division did a really poor job on the color math, and it didn't work out. (Things have improved a lot since then, think Berns and Taplin).</p>

<p>Surprisingly, there's little to no effect on noise or dynamic range. You often see reasoning like this.</p>

<blockquote>

<p>Just using current technology, typical digicams have pixel density of around 50MP/sq.cm. That translates to an APS sensor with 165MP or a full frame sensor with 432MP - but with the noise performance of a P&S digicam.</p>

</blockquote>

<p>But that ignores the fact that the goal isn't a 12mp crop out of the 432mp FF image, it's an image using all 432mp, and often at the same size as the P&S image. Humans aren't growing larger at a "Moore's Law" rate, 8x10 is still a comfortable print to hold, 11x17 is still a tabloid. Print an ISO 1600 image from a 24mp FF like a Nikon D3X and a 12mp FF like a D3 at the same size (8x10, 11x17, your choice) and they look surprisingly similar in noise character. It's only the pixel peepers who say you can't shoot the D3X at ISO 1600 or 3200. By the same token, if you take a 49 shot 7x7 panaorama on a 12mp P&S and stitch it to 430mp and print at 11x17, the dynamic range and shadow detail are going to surprise you.</p>

<blockquote>

<p>while 3 (big) CCDs always wins in color cameras.</p>

</blockquote>

<p>That is only true if you redefine "win" as "lose big time". 3CCD is a one-trick pony from the early days of video cameras. It gives you the highest possible photon efficiency: the dichroic coated prisms insure that pretty much every photon hits one of the 3 sensors and is detected. But you pay a terrible price for that:</p>

<ul>

<li>Horrible color accuracy: really only good enough for vector color (NTSC, PAL, or SECAM) video. It's basically a tool for reportage, where terrible color is an acceptable price for seeing the tear on a mother's face in a news cast.</li>

<li>Optical compromises: because you've got 2 solid glass prisms in the optical path, your "back focus", the distance from the rear element of the lens to the focal plane, is on the order of three image heights, say 75mm on a FF sensor. That limits the focal length for a symmetrical f2.8 (the "core" of wide angle designs) to about 85mm, so a simple 24mm wide is now a large, heavy 85mm f2.8 with 3 stages of reverse Galilean "wide converter" stacked in front of it, to get from 85mm to 24mm. Huge, expensive, heavy, and low performing compared to a 24mm f2.8 for an existing 39mm back focus FF SLR, let alone a mirrorless camera like a Leica M9.</li>

<li>Lack of focal plane shutters. Any 3CCD camera has to either have an in-lens leaf shutter, a poorly performing behind-lens leaf shutter, or a sensor based "snap shutter". That's fine for video, but in still photography, remember the Nikon D70? The snap shutter came at the price of compromised shadow detail and a big blooming problem.</li>

</ul>

<p> My wife said she does not want one around the house. She thinks I would be better off buying a paint brush and a gallon of paint.</p>

<p>The day we get a 1GB full frame camera (or even a 100GB camera) is the day before (or maybe the minute before) the forums will be full of complaints about how bad even the most expensive lenses are and how inaccurate autofocus is.</p>

<p>Of course we get a few of those now, but we'll get more of them.</p>

<p>I guess we are approaching the point where the light sensitive element will be at the molecular scale. Hang on, that'll be just like film!</p>

<p>It's as academic as it was in the age of film. How many of us were making contact prints from Banquet cameras? Enlargements from 8x10s? 4x5s? How many actually needed those formats?</p>

<p>Right tool for the job. And in the not-too distant future, pixels will be a thing of the past, anyway.</p>

<blockquote>

<p>many others will convince themselves that they just can't do without 1GP even if they just do web uploads 1000 px wide.</p>

</blockquote>

<p>On another forum I used to participate in there was a question about how many people actually print their images. 80% of them just put them on websites or viewed them on their monitors but these were the same people who were always buying newer, higher resolution cameras.</p>

 

<blockquote>

<p>Enlargements from 8x10s? 4x5s? How many actually needed those formats?</p>

</blockquote>

<p>Before digital existed, most amateurs used 35mm and were happy with 4x6 or 5x7 prints. When digital arrived and it was easier to print at home, everyone suddenly had to have all of their great works of art printed as large as their printer would allow.</p>

 

<blockquote>

<p>And in the not-too distant future, pixels will be a thing of the past, anyway.</p>

</blockquote>

<p>Care to elaborate?</p>

<p>Thanks to all for some enlightening viewpoints on this question, and to those who have yet to comment. Seeking high image quality may be overkill in these days of Internet image presentation and the 4 x 6 or 5 x 7 inch print that most seek (with perhaps the occasional 11 x 14 inch print that we can proudly display in our living area, but this is similar to the needs of the public in pre-digital days).</p>

<p>However, those who decide to spend hard earned dollars on a 16 x 20 inch, or larger, photograph that is displayed in a gallery or other site merit that a particular attention has been paid to image resolution, image tone, colour quality and cleanlinesss, as well as print quality and permanence. While the print buyers at our small gallery (primarily tourists rather than art collectors, per se) are more concerned with the image content, and how they react to it, than most of these parameters, I am incited to write into the accompanying authenticity certificate the details of the ink and print materials used, as well as any archival treatment that a wet print has received. The quality of the tonality and resolution of the image may not be quantified but I believe these are things that may well become more apparent to the viewer upon repeated personal contact with the print.</p>

<p>Going that extra distance in achieving a quality image from the detail resolution and chomatic standpoints has some importance and is not that easy to assure with the small sensor size cameras many can afford to use. The limiting weakest link in all this for small camera users may be sensor resolution, but it may also be pixel size, camera firmware or lens performance, not to mention the ability of the photographer (use or not of a tripod, post exposure manipulation, etc.). Unit image cost, camera maniability and operational ability of the photographer are limitations on the use of the whole plate, 8 x 10 and 11 x 14 film formats, although the results of those photographers often establish the upper limits of print quality that might be attainable with the smaller digital camera (although recognising that this does not seem to be the primary objective of the industry or its clients).</p>

<p> </p>

<blockquote>

<p>Enlargements from 8x10s? 4x5s? How many actually needed those formats?</p>

</blockquote>

<p>Oh, me, me!</p>

<p>My 4x5 enlarger and camera were homebrew, but played really well together. ;) Slow 4x5 had such a lovely tonality on an 8x10 or 10x14. Flower macros just had a wonderful feeling that's hard to recapture digitally.</p>

<p>it seems to me, for those printing max size at home, 36x24 inches, and demanding the highest, extreme DPI for such a large print (300 dpi); the max sensors for a "35mm" 3x2 format need to peak at 10800x7200 pixels, or 77 MP. As long as the top lenses can resolve at such a fine frequency then Bring it On! (the year 2016 or so, no doubt).</p>