Bill Pierce
Well-known
I took a break today and browsed a number of photo discussion web sites. There seems to be a lot of misinformation about sensor size and pixel counts.
All other things being equal (and they’re not because sensor technology continues to advance) larger pixels have higher signal to noise ratios.
All other things being equal, more pixels mean larger prints without a loss of definition.
Bigger pixels - less noise. More pixels - bigger prints.
Thus, a Canon S90 or 95 with its 10 megapixel sensor, shows less noise at high ISO’s than the same sized sensor in the G10 that crams 15 megapixels into the same area. But, use the G10 at its lowest ISO of 80 and you can make bigger prints that still hold detail than you can with the S90 or 95.
How do you get the best of both worlds - big prints of dimly lit subjects shot at high ISO’s? You use a big camera with a big CMOS sensor that has big pixels and lots of them, probably a full frame Nikon or Canon. And, although progress seems a little slower in the big full-framers than the little fellows, expect all your current cameras to be outdated before you are.
(And if that makes you uncomfortable, just think what will happen if Leitz and Kodak come up with CCD sensor that can run at high ISO’s or Sigma actually markets a camera with a big Foveon.)
OK, that's the weekend rant. I'm not quite sure how one replies to it.
All other things being equal (and they’re not because sensor technology continues to advance) larger pixels have higher signal to noise ratios.
All other things being equal, more pixels mean larger prints without a loss of definition.
Bigger pixels - less noise. More pixels - bigger prints.
Thus, a Canon S90 or 95 with its 10 megapixel sensor, shows less noise at high ISO’s than the same sized sensor in the G10 that crams 15 megapixels into the same area. But, use the G10 at its lowest ISO of 80 and you can make bigger prints that still hold detail than you can with the S90 or 95.
How do you get the best of both worlds - big prints of dimly lit subjects shot at high ISO’s? You use a big camera with a big CMOS sensor that has big pixels and lots of them, probably a full frame Nikon or Canon. And, although progress seems a little slower in the big full-framers than the little fellows, expect all your current cameras to be outdated before you are.
(And if that makes you uncomfortable, just think what will happen if Leitz and Kodak come up with CCD sensor that can run at high ISO’s or Sigma actually markets a camera with a big Foveon.)
OK, that's the weekend rant. I'm not quite sure how one replies to it.
LKeithR
Improving daily--I think.
Pixel mania has been around pretty much since the first digi-cam hit the market. Pixel count is, for sure, an important measure of a cameras performance but, in my view at least, it is only one of several. Some people seem to obsess over pixel counts and pixel peeping; some don't. I tend to be one of the latter...
tlitody
Well-known
There comes a point where you don't need anymore pixels and for most purposes that point has already been reached which is why 12MP sell like hot cakes even for top pros when double that pixel count is available at the same price. Only when you need the ultimate do you need to go to MF sensor sizes and the price of that kit makes it impossible for all but the top pros. So the useful life of a digital camera is now much extended compared to a few years ago. Buy one today and there is no reason why it won't be good enough in 10 years regardless of how many pixels you can have in 10 years.
Same with PC's. How many Giga Hz speed do you need? How much memory do you need? The answer is you only need the latest and greatest if you want to run the latest and greatest software. But photography is not about software and PS CS is still good enough for today.
There are limits to requirements. For example, I am writing this on a 400Mhz Pentium II PC which is configured as a web server running apache, mysql, php and hosts my web site from home. I do all my PC work and web design on it. It's only become a pain for some stuff recently and will get upgraded soon. I've avoided all the incremental upgrades that the gas afflicted waste money on. If I wanted to run HD video I would have had to upgrade much sooner but I haven't bought into needing HD video on PC so I have never needed more processing power.
Same with PC's. How many Giga Hz speed do you need? How much memory do you need? The answer is you only need the latest and greatest if you want to run the latest and greatest software. But photography is not about software and PS CS is still good enough for today.
There are limits to requirements. For example, I am writing this on a 400Mhz Pentium II PC which is configured as a web server running apache, mysql, php and hosts my web site from home. I do all my PC work and web design on it. It's only become a pain for some stuff recently and will get upgraded soon. I've avoided all the incremental upgrades that the gas afflicted waste money on. If I wanted to run HD video I would have had to upgrade much sooner but I haven't bought into needing HD video on PC so I have never needed more processing power.
Chris101
summicronia
Isn't this part of the commonly accepted knowledge about digital cameras by now?
willie_901
Veteran
Emil Martinec wrote a detailed technical article Noise, Dynamic Range and Bit Depth in Digital SLRs about current digital sensor technology. His technical article covers the physics of sensor design and shows how empirical measurements agree with theory.
In digital photography we estimate how many photons interact with a particular sensor site. This is the only data we record. Martinec focuses mostly on noise because noise, or uncertainty in parameter estimation (photon counts), is always undesirable.
On page three of his article Martinec writes:
Bottom line: Among the important measures of image quality are signal-to-noise ratio of the capture process, and resolution. It was shown that for fixed sensor format, the light collection efficiency per unit area is essentially independent of pixel size, over a huge range of pixel sizes from 2 microns to over 8 microns, and is therefore independent of the number of megapixels. Noise performance per unit area was seen to be only weakly dependent on pixel size. The S/N ratio per unit area is much the same over a wide range of pixel sizes. There is an advantage to big pixels in low light (high ISO) applications, where read noise is an important detractor from image quality, and big pixels currently have lower read noise than aggregations of small pixels of equal area. For low ISO applications, the situation is reversed in current implementations -- if anything, smaller pixels perform somewhat better in terms of S/N ratio (while offering more resolution). [....] Rather than having strong dependence on the pixel size, the noise performance instead depends quite strongly on sensor size -- bigger sensors yield higher quality images, by capturing more signal (photons).
The other main measure of image quality is the resolution in line pairs/picture height; it is by definition independent of the sensor size, and depends only on the megapixel count. The more megapixels, the more resolution, up to the limits imposed by the system's optics.
That's all there is to it.
For the best image quality (least uncertainty in photon count estimates) you want the largest sensor you can afford. If you prefer the highest possible resolution, then you want pixel density. If you want the best low-light performance you choose a large sensor with large pixels.
Martinec discusses ISO selection as well:
Somewhat counter-intuitively, for fixed aperture/shutter speed, it is best to use the highest possible ISO (without clipping highlights); this result is consistent with the ETTR [expose to the right] philosophy, since using higher ISO pushes the histogram to the right if one thinks about things in terms of raw levels (ADU) [RAW unit values]. However, the benefit from the use of higher ISO comes in the shadows, not in the highlights where "there are more levels"; to demonstrate that will require some more detailed analysis of the noise and S/N graphs on page 2.
[....]
Bottom line: Read noise at high ISO is much smaller than read noise at low ISO, in terms of the error in photon counting that it represents. Thus, better image quality is obtained for using the highest ISO for which the signal is not clipped.
So high ISO performance is not just a marketing fad. Auto ISO is beneficial (assuming the automation algorithm chooses the highest ISO) if the photographer takes care to avoid over exposure.
A lot can happen after RAW data leaves the sensor and arrives in the memory card. So Martinec's technical analysis is just one part of the story.
Image content is much more important than image quality (as long as a low quality level does not distract from the content). Modern digital cameras most RFF readers would consider produce image qualities that do not interfere with content.
In digital photography we estimate how many photons interact with a particular sensor site. This is the only data we record. Martinec focuses mostly on noise because noise, or uncertainty in parameter estimation (photon counts), is always undesirable.
On page three of his article Martinec writes:
Bottom line: Among the important measures of image quality are signal-to-noise ratio of the capture process, and resolution. It was shown that for fixed sensor format, the light collection efficiency per unit area is essentially independent of pixel size, over a huge range of pixel sizes from 2 microns to over 8 microns, and is therefore independent of the number of megapixels. Noise performance per unit area was seen to be only weakly dependent on pixel size. The S/N ratio per unit area is much the same over a wide range of pixel sizes. There is an advantage to big pixels in low light (high ISO) applications, where read noise is an important detractor from image quality, and big pixels currently have lower read noise than aggregations of small pixels of equal area. For low ISO applications, the situation is reversed in current implementations -- if anything, smaller pixels perform somewhat better in terms of S/N ratio (while offering more resolution). [....] Rather than having strong dependence on the pixel size, the noise performance instead depends quite strongly on sensor size -- bigger sensors yield higher quality images, by capturing more signal (photons).
The other main measure of image quality is the resolution in line pairs/picture height; it is by definition independent of the sensor size, and depends only on the megapixel count. The more megapixels, the more resolution, up to the limits imposed by the system's optics.
That's all there is to it.
For the best image quality (least uncertainty in photon count estimates) you want the largest sensor you can afford. If you prefer the highest possible resolution, then you want pixel density. If you want the best low-light performance you choose a large sensor with large pixels.
Martinec discusses ISO selection as well:
Somewhat counter-intuitively, for fixed aperture/shutter speed, it is best to use the highest possible ISO (without clipping highlights); this result is consistent with the ETTR [expose to the right] philosophy, since using higher ISO pushes the histogram to the right if one thinks about things in terms of raw levels (ADU) [RAW unit values]. However, the benefit from the use of higher ISO comes in the shadows, not in the highlights where "there are more levels"; to demonstrate that will require some more detailed analysis of the noise and S/N graphs on page 2.
[....]
Bottom line: Read noise at high ISO is much smaller than read noise at low ISO, in terms of the error in photon counting that it represents. Thus, better image quality is obtained for using the highest ISO for which the signal is not clipped.
So high ISO performance is not just a marketing fad. Auto ISO is beneficial (assuming the automation algorithm chooses the highest ISO) if the photographer takes care to avoid over exposure.
A lot can happen after RAW data leaves the sensor and arrives in the memory card. So Martinec's technical analysis is just one part of the story.
Image content is much more important than image quality (as long as a low quality level does not distract from the content). Modern digital cameras most RFF readers would consider produce image qualities that do not interfere with content.
charjohncarter
Veteran
Right now, the only way to achieve both your objectives is to buy a Leica S2, a Pentax 645D, a Hassy with D-back, or any of the MF digitals. If you don't want to do that, buy whatever suits your purposes.
sojournerphoto
Veteran
Maybe, but it can't add resolution that isn't there. All it achieves is to spread out the real information in a pretty way.
Matus
Well-known
Somehow it is a simple math and requirements one has - at least for me.
If I want to be able to print say 16x20" with full detail (at 360 dpi, for example) that means that I need 5760 x 7200 (or 40 Mpixels) pixels to do so. If I can not afford a camera with that resolution I need to check whether less (say 240) dpi will give me the result (print) I want - most probably it will. Then I am at about 20 Mpixels - what is indeed doable. Of course one just shoot medium format (or larger) film and get there too. 16x20" from 4x5" is hard to beat
Now - if I need (for whatever reason) good high ISO performance than I have to see what I can get there. In particular with BW the digital noise can be more acceptable.
One can argue that with the exception of few Sigma cameras that are using Foveon technology - we all use and are used to Bayer interpolated images what also means that watching the digital files at 100% just can NOT look perfectly sharp and detailed. This however makes (nearly) no difference in a print - you simply print at that dpi that you find good enough. The images produced with Foveon sensors can handle some decent amount of upsizing until they look as "bad" as Bayer interpolated images.
So - with the above one can define his/her needs and choose a camera accordingly. Of course - for many type of photos the level of detail is not so relevant so what is mentioned above does not necessarily apply and different (even if still technical) points will matter.
It is all just matter of getting a tool (camera in this case) that is capable of the work you want to do. That is the easy part of getting the image you want.
I personally would love to see in the future Foveon sensors replacing bayer interpolation. I like the idea that the data stored are unnecessarily large what makes storage and processing easier. Whether that will ever happen I do not know.
In the above I have completely ignored the content of the photo. Now - that does say something about me ...
If I want to be able to print say 16x20" with full detail (at 360 dpi, for example) that means that I need 5760 x 7200 (or 40 Mpixels) pixels to do so. If I can not afford a camera with that resolution I need to check whether less (say 240) dpi will give me the result (print) I want - most probably it will. Then I am at about 20 Mpixels - what is indeed doable. Of course one just shoot medium format (or larger) film and get there too. 16x20" from 4x5" is hard to beat
Now - if I need (for whatever reason) good high ISO performance than I have to see what I can get there. In particular with BW the digital noise can be more acceptable.
One can argue that with the exception of few Sigma cameras that are using Foveon technology - we all use and are used to Bayer interpolated images what also means that watching the digital files at 100% just can NOT look perfectly sharp and detailed. This however makes (nearly) no difference in a print - you simply print at that dpi that you find good enough. The images produced with Foveon sensors can handle some decent amount of upsizing until they look as "bad" as Bayer interpolated images.
So - with the above one can define his/her needs and choose a camera accordingly. Of course - for many type of photos the level of detail is not so relevant so what is mentioned above does not necessarily apply and different (even if still technical) points will matter.
It is all just matter of getting a tool (camera in this case) that is capable of the work you want to do. That is the easy part of getting the image you want.
I personally would love to see in the future Foveon sensors replacing bayer interpolation. I like the idea that the data stored are unnecessarily large what makes storage and processing easier. Whether that will ever happen I do not know.
In the above I have completely ignored the content of the photo. Now - that does say something about me ...
tlitody
Well-known
if future sensors are going to be so much better then what we have now must be really useless. So why do nearly all pros use them? Why aren't they all using film? Answer is simple, because what we have now is plenty good enough. Don't worry about what is to come because it won't be visibly any better unless you are a pixel peeper. And as digital viewing becomes increasingly dominant where we only have low resolution screens, then there will virtually zero advantage in increased pixel counts in cameras. Sensor pixel counts are already way bigger than your average display unit.
sol33
Established
Thanks Willie, that was an interesting post.
Gabriel M.A.
My Red Dot Glows For You
Did you mean to type "larger pixels have lower signal to noise ratios"? Which, btw, I think what this meant to say is that the bigger the sensor, given a specific sensor/area density, the less the perceived noise.
I know that there was a complaint about the image quality between the Canon 40D and the Canon 50D: the 50D has a higher megapixel count than the 40D, and the 40D is perceived to have less high ISO noise, and reduced dynamic range.
To make things even more complicated in this bumper-sticker soundbite Twitter world, the Canon 10D, for example, is by far cleaner at low and mid ISO sensitivities, but luma and chroma noise in the Canon 10D in ISO 100-400 is far far far far less (I'd say imperceptible at ISO 100) than the same ISO ratings in the newer (and hence, "better" for most people) models.
Except Nikon, of course. Those Nikon cameras are flawless and are God's gift to men
Gabriel M.A.
My Red Dot Glows For You
tlitody
Well-known
No I think he said it correctly. The larger the ratio of signal to noise the less noise there is relative to signal. i.e. 1000:1 is better than 100:1 cos 1000:1 means 0.1% noise whereas 100:1 means 1% noise.
shadowfox
Darkroom printing lives
Bill that reminds me, I never quite figure out why Sigma has not come out with a full-frame foveon camera?
Wouldn't that be a guaranteed head-turner among professionals and serious enthusiasts?
Instead they kept with their SD line which is neither here nor there. Yes, I have used SD14, it's different and not necessarily in a good way.
Rob-F
Likes Leicas
While we are on this subject, it occurs to me that if Leica had made the M9 with a 12MP sensor, it would probably have all the pixels many folks need; and the high-ISO performance could have been better.
Roberto V.
Le surrèalisme, c'est moi
Related story:
One time I went to a yearly road trip wit some of my closest friends. I took my Canon 10D with a 50mm 1.8 and a 28-105mm. One afternoon, we were going to take a group photo at night, in a house that my friend has in a small town. I took my 10D and was setting the timer, when my best friend says "cool camera, how many megapixels does it have?" I told him it had 6.3, to which he replied "this one has 8, I think we should use it instead", pointing to his conmpact Sony P&S
:bang:
One time I went to a yearly road trip wit some of my closest friends. I took my Canon 10D with a 50mm 1.8 and a 28-105mm. One afternoon, we were going to take a group photo at night, in a house that my friend has in a small town. I took my 10D and was setting the timer, when my best friend says "cool camera, how many megapixels does it have?" I told him it had 6.3, to which he replied "this one has 8, I think we should use it instead", pointing to his conmpact Sony P&S
:bang:
It's Conservation of Inconvenience. I first heard that phrase in 1979 applied in the field of X-Ray optics. Holds true in pretty much anything.