Pixel Density / Pixel Size Question

I am a bit confused by some of the comments I’ve read concerning pixel density and/or pixel size. For the purposes of this discussion, let’s assume the following:

Cameras:
1. Canon 20D, 8.2 MP (3504 x 2336), 22.5mm x 15.0mm sensor, 24253 pixels/mm^2
2. Canon 5D, 12.8 MP (4368 x 2912), 35.8mm x 23.9mm sensor, 14866 pixels/mm^2

Also, any object being photographed fills each frame equally (either by using a longer lens on the 5D or simply moving closer). In other words, crop factor is not part of the equation.

Under these circumstances, I’m having a hard time grasping how having a smaller, lower-resolution sensor with a higher pixel density could ever be an advantage (other then being able to be used in smaller, lighter cameras). My understanding is that smaller pixels collect less light (information) and display higher levels of noise. (I also understand that noise reduction built into the cameras help to lessen the noise issue.)

I have read comments that with a denser 1.6x sensor (and smaller pixels), you get more pixels on the subject and capture better detail. How is that possible if the subject fills both frames, especially when the 1x sensor has an extra 4 MP? If the object fills both frames, the 5D sensor would actually have more pixels on the subject wouldn’t it? I could see the crop factor being an issue if you couldn’t use a longer lens or get closer to the subject, but that is an altogether different issue in my mind.

If anyone could enlighten me on the pros and cons of pixel size and density, I would greatly appreciate it. (Once again, I understand the crop factor issue and I know that the 20D has a faster frame rate than the 5D. I'm not arguing the pros and cons of these cameras. I just want to make sure that I understand the sensor issues independent of these other factors.)

BTBeilke,

I think you have a perfect understanding of this. As you said "all things being equal", as in frameing the subject the same ratio of the image size, the FF sensor would likely have less noise and therefore give a better picture. Some people would argue that the center of a lens is of higher quality than the edges and therefore, using a 1.6 sensor you "crop out" the bad quality light at the edges and have a crisper image. So based on this, it is a matter of preference on noise vs. "image quality" (if you agree with these rationalizations of using smaller frame sensors with large frame lenses).

When you get rid of the "all things being equal" comment, that is where the arguement gets on some terms that we can quantify more than someones impression of quality. If you were into landscape and other wide-angle photography, the FF sensor would be better because you get a bigger FOV with the same lens. The bigger the FOV the lens has, generally the more expensive it is, so for landscapes a FF gives you better "bang for the buck".

In the same way, a 1.6 sensor gives better value when you consider the telephoto end. Suppose I wanted to take a picture of the moon, but all I have in my lens collection is the 70-200mm f2.8 (and this is the case for me. ), the small frame sensor will give me more detail in the moon than a FF sensor. This is because the 200mm end of my lens is always going to project the moon at a certain size onto a sensor (no matter if it is a FF or 1.6). With the FF sensor, I'm just going to be cropping away more blackness of space. This blackness is contributing nothing to my image detail. But, on the 1.6 sensor, the projected image of the moon has a much higher pixel density associated with it and therefore I can get more detail out of the picture with using the same lens.

Really, it boils down to what type of photography you enjoy and how much money you can throw at the problem. I could just as easily buy a 100-400mm lens for a FF camera and give a similar (probably better) quality picture out of my moon problem.

Hope that helps (and anyone else feel free to comment if I have things entirely wrong)...

Patrick

Both right
I've come to think of it as an issue of pixels/FoV. For a given FoV , as in the moon with a 200mm above, the more pixels you can capture the image with, the better.
BTB : "(either by using a longer lens on the 5D or simply moving closer)" - this is often not possible, especially for bird (small subject) photography. 1) you may not be able to afford the longer lens 2) there may not BE a longer lens and 3) move closer and the subject flies away. This is why birders like the 20D - it is Canon's leader in the MP/FoV category. If you can use all 12 MP to make your image, then clearly you should.

Thanks for the replies.

I understand (I think) the pros/cons of the crop factor issue and certainly there will be times when you can't get close enough with a FF camera to make up for the 1.6x crop factor. I was just trying to separate the quality issues inherent to the sensor from these other factors.

BTW, according to my calcs, if you were to crop a 5D image down to 20D, the effective crop factor would be 1.247x. So, you gain some of the difference back due simply to the higher resolution of the 5D.

Not sure what that means or how you calculated it!

Maybe this is the converse : to get the same pixel density as a 20D, a "full frame" sensor would need ~21MP

That is really confusing me too - good questtion!

I'm not sure I'm clear though even after the answers

Another way to think of this is by analogy to film - more-closely spaced pixels are like finer-grained film - they can capture finer details (e.g. more lp/mm on a test-chart).

The 20 D uses a smaller piece of film but it's "finer-grained". That film is not yet widely available in full frame size, only APS-C aka 1.6x size. When a Canon 22MP FF sensor is released, it will be the same "film" that is used in the 10D

That's all it needed - an analogy with Film - NOW it get it

I am really confused ?? so which is better?

Better for what purpose???

BTW, you are exactly right about needing an almost 21MP FF image in order to crop down to 20D resolution and still have an equivalent to the 1.6x crop factor.

However, I don't see how the comparison to film is a good analogy at all. These are vastly different technologies. In fact, if you want to use a film analogy, wouldn't you compare noise since that is the digital equivalent to grain? In that case, the 20D would be the noisier/grainier film.

When I open up a digital image, a pixel is a pixel. They are all exactly the same size regardless of the size of the photosite that captured them. I keep hearing people (20D owners?) claiming that the 20D captures greater detail. If you go back to the assumptions at the begining of the post (i.e. the subject fills both frames equally) I don't see how this could be possible. For starters, you would be getting a higher resolution picture with more pixels defining the subject with the 5D. Further, how could a smaller, and therefore less sensitive, photosite possibly collect more information for any one pixel than a larger photosite that collects more light (information) and has a better signal-to-noise ratio?

Now, if you can't use a longer lens or get closer to the subject, the reverse would be true. You'd end up with more pixels on the subject (more detail) with the 20D than you would with the 5D due to the 20D's 1.6x crop factor. But that is a far different issue than saying that the 20D sensor has some inherent detail advantage over the 5D. That is a 1.6x crop factor advantage, not a sensor technology advantage.

I know this for sure. In every data collection activity in which I've been involved, a larger data set has always meant more accurate results. It would seem that the 5D has a definite advantage in that department. That is not to say that it is the best camera for every situation. In fact, I am debating which of these cameras to buy which is difficult because they are both so good. But I'm just not seeing how it can be argued that the 20D sensor technology is better than the 5D's.

go back to the original post!

Thanks Jerry, very helpful, not!

The original post asked for enlightenment for a specific set of assumptions. Not for a blanket statement as to which is "better"!

Since then some other conditions have been discussed, so the meaning of a "Which is better?" question isn't so obvious!

From the original post, where the assumption was that the composition is the same (because the cameras use different lenses), the 12MP option is clearly capturing more details than the 8MP option - in that sense it is "better". That was already accepted in the 3rd post

The enlightenment was around the issue of when you can't fill the frame with your longest lens with the 1.6 camera (common with photography of small critters). In that case, the 8MP option is "better" because - if you can't fill the frame of a 1.6 you surely can't fill the frame of a FF camera - the assumption (about equal framing) breaks down.

BTB - folks who feel the film analogy is relevant to a "detail" discussion are familiar with the concept of fine-grain ( dense pixel spacing) and coarse grain (large pixel spacing) films. When I scan my film (Provia100F) at 2500 ppi, I don't see grain to any significant extent. When I scanned NPS160, a Fuji wide dynamic range negative color film, it's terribly grainy and I can't print it as large without the grain being very evident, compared to the Provia at the same enlargement. In any case, it was "only" an analogy and certainly noise plays a role in measurements of resolution as do MTFs etc.

BTB said:
"Now, if you can't use a longer lens or get closer to the subject, the reverse would be true. You'd end up with more pixels on the subject (more detail) with the 20D than you would with the 5D due to the 20D's 1.6x crop factor. But that is a far different issue than saying that the 20D sensor has some inherent detail advantage over the 5D. That is a 1.6x crop factor advantage, not a sensor technology advantage."

I am indeed saying that the 20D sensor has some inherent detail advantage over the sensor in the 5D, because it has a higher pixel density, linear resolution, Nyquist frequency (however you want to name it). This is referring to the "quality" not "quantity" of the details. Look at the central 10mm by 10mm of the 20D, the 5D and the mythical 1Ds3 (the one which has 22MP in a FF sensor) when e.g. a 500 mm lens is focused on a small bird, whose image falling on the sensor is 9mm high. The 20D and "1Ds3" will produce an image of the bird that is 1400 pixels high, while the 5D image is 1100 pixels high. (In terms of data collection - I recently retired from biophysics and spectroscopy so I'm there - this is substantially higher density data, so finer detail is recorded) This is *solely* due to the pixel spacing and *nothing* to do with the format or crop factor. The total number of pixels captured by the three cameras is different (8, 12 and 22) but the amount of detail that can be made out on the bird's feathers is the same for the 8 and 22 and it's more than from the 12. That's all I'm saying.

AJSJones wrote:
I am indeed saying that the 20D sensor has some inherent detail advantage over the sensor in the 5D, because it has a higher pixel density, linear resolution, Nyquist frequency (however you want to name it). This is referring to the "quality" not "quantity" of the details. Look at the central 10mm by 10mm of the 20D, the 5D and the mythical 1Ds3 (the one which has 22MP in a FF sensor) when e.g. a 500 mm lens is focused on a small bird, whose image falling on the sensor is 9mm high. The 20D and "1Ds3" will produce an image of the bird that is 1400 pixels high, while the 5D image is 1100 pixels high. (In terms of data collection - I recently retired from biophysics and spectroscopy so I'm there - this is substantially higher density data, so finer detail is recorded) This is *solely* due to the pixel spacing and *nothing* to do with the format or crop factor.

Well, I think we've already established that the 20D has an advantage over the 5D in telephoto applications when you limit both cameras to the same distance and focal length (as in your example). Those were exactly the kinds of limitations I was trying to eliminate from this discussion. And in this specific scenario, it is probably technically accurate to say that it has nothing to do with the crop factor. But I don't see how you can really separate it out in any practical manner. After all, if the 20D didn't have a smaller APS-sized sensor (and therefore a 1.6x crop factor) onto which they made compromises in order to cram more photosites, it wouldn't have a higher pixel density in the first place.

To adapt your example to my original question, which camera would create the better picture (detail, noise, dynamic range, etc.) if the bird was framed to capture an image 1400 pixels tall on each camera?

The real heart of my original question could be restated like this: If you take a single photosite/pixel of the 5D sensor and a single photosite/pixel of the 20D sensor, which one would capture more accurate information for that single pixel? I don't see how it would be possible for the smaller photosite to be better. But, if I'm wrong, I'd love it if you could explain the science to me. (I'm an engineer so I like the technical/mathematical aspects of this discussion.)

Based on your background, here is another related question that comes to mind. Let's say that you were asked by the military to create the highest quality 100MP CMOS sensor that money could buy. Let's also assume that there were no limitations on the size/weight/cost of the body or lenses that would be used with this sensor. Image quality is all that matters. Would you purposely create a small sensor with tiny photosites and an incredibly high (and noisy?) pixel density or a larger sensor with larger photosites?

For the record, I'm not trying to diminish the 20D at all. That may very well be the camera that I end up buying. I just want to make sure I understand the sensor technology independent of the camera as best I can. In the past I've read discussions about a 12-16MP APS sensor where everyone was concerned about noise and image quality due to the small pixels and higher pixel density. Recently, it seems like a lot of 20D users/proponents are singing a vastly different tune and I'd just like to cut through the smoke and get to the facts.

Thanks for the discussion!

I separate it out in a practical manner when I take pictures of birds with my longest glass as close as I can get and they don't fill my VF. The quality of the noise is not an issue until you get to quite high ISO like 1600.
All other things being equal, a larger or higher photon capacity, sensor element will have better noise characteristics and therefore a larger dynamic range. If that was your question initally, there'd have been no discussion I don't think anyone said the 20D pixels were "better" than those in the upcoming 5D, they've said the 20D sensor has higher linear resolution than their expectation of that from the 5D (or other sensors with wider pixel spacing) and that they like that. I do and it has practical consequences. I don't think we're in disagreement, are we?

Now, the 1Ds has 11MP in a FF format but as I understand it, it sucks at high ISO. Digic2 chip technology has given the 20D pixels much better S/N than those of the 1Ds even though they are significantly smaller, so all things are not necessarily equal. The facts you are seeking are, I suspect, changing as we discuss them.

Your 100MP CMOS sensor would have to have some design criteria to define "highest quality". Check out slide 26 of this pdf www.bnl.gov/henp/docs/​doe_hep04/slides_arons​on2.pdf (cut and paste - I didn't want it to download into the post !) It's a 2.8 Gigapixel array and it uses some neat lenses - I'd say it's pretty high resolution. Define the "noise " level you think is best "quality" under some range of conditions and we can look at the available sensor design data and we can start designing!