Pixel density and noise?

As my tag line says, I continue to agonize on the 50D vs 5D2 decision. I have read much about them, and focused on the 1.6 sensor vs FF, and the pros and cons of each. Because I am big on avoiding noise, there are many factors such as ISO, etc. But here is a question I don't have a good answer for . . .

Is pixel density a factor that I should consider, and what are the pros and cons of the different values?

The 5D2 has a pixel density of .0255MP per square mm, and the 50D has a density of .0466MP per mm. (number of megapixels divided by the sensor square area in mm - Thanks Tony for the math catch). My question is, would a denser number of pixels be better or worse? I can generate my own speculation for both directions, but I am curious as to what more experienced folks have to say.

What say you? Am I just chasing esoteric data, or am I looking at an objective measurement criteria??

Thanks all . . .

14.67 MP / mm^2?? Your arithmetic is way off there.

50D sensor: 22.3x14.9mm, 4752x3168 (332.27 mm^2, 15054336 pixels = 45307/mm^2)
5DmkII sensor: 24.0x36.0 mm, 5616 x 3744( 864 mm^2, 21026304 pixels = 24336/mm^2)

Note that the 50D's sensor's sites are gapless while the 5DmkII still has some gaps, perhaps so that the grid can be varied at the corners . Whether higher density is desirable depends on one's needs. In general, larger pixel sites tend to give lower noise, all other factors being equal.

Math corrected - Thanks Tony . . .

In actual photographs, the 5D2 is better. Less magnification to any given final display is what gives the 5D2 its ultimate edge.

Like Anthony said, larger pixel sites yield a higher signal-to-noise ratio -- that is, on a per-pixel basis they will have the same or more noise but will collect more light-per-noise.

Newer generations of sensors have improved the noise levels, both with the smaller sensors and the larger sensors, but still if you compare the older 5D Classic with the newer bodies with smaller sensors and greater pixel densities, the larger sensor with the larger pixel wells wins for cleanness and all that entails.

But, with the larger sensor bodies there are tradeoffs: higher prices, and the smaller sensors provide some performance benefits as well as the ability to put more pixels in the center of the "image circle" so that if you needed to crop a full frame image to the size of the smaller sensor image the smaller sensors would have greater resolution, although with the noise factor.

Each photog has to weigh everything into the choice, or eventually have one of each!

You know, instead of agonizing over specs (although pixel density is often a good indicator of relative per-pixel noise performance, all else being equal) you could just look at sample photos from both and see if the 5DII's cleaner output is worth the extra money for your use.

golfecho wrote in post #7749915
Is pixel density a factor that I should consider

Yes, you should consider the effect of pixel density on:

• File size and workflow
  
• Magnification value
  
• Out-of-camera JPEG

golfecho wrote in post #7749915
What are the pros and cons of the different values?

Higher pixel densities require bigger files, slower workflow, longer processing times, higher magnification for telephoto/macro. For me this is not a factor, but it may be important to some shooters. Lower pixel densities result in smaller files, faster workflow, and lower magnification.

golfecho wrote in post #7749915
The 5D2 has a pixel density of .0255MP per square mm, and the 50D has a density of .0466MP per mm.

They also have a hugely different sensor size. It makes no sense to analyze the impact of pixel density without also considering the sensor size.

A digicam-sized sensor with super-large pixels (0.21 MP) is never going to be superior to a FF35 sensor with super-tiny pixels (21 MP).

anthony11 wrote in post #7750167
In general, larger pixel sites tend to give lower noise, all other factors being equal.

I kindly disagree. Larger pixels only give lower noise *per pixel* AOTBE. Looking at it that way is not seeing the forest for the trees. Larger trees doesn't mean more wood in the forest. You have to also consider the number of trees to know how much boardfeet is contained in the entire forest.

tonylong wrote in post #7750536
Like Anthony said, larger pixel sites yield a higher signal-to-noise ratio -- that is, on a per-pixel basis they will have the same or more noise but will collect more light-per-noise.

The per-pixel basis doesn't tell us anything about the image as a whole. See the following explanation.

Noise scales with spatial frequency

tonylong wrote in post #7750536
if you compare the older 5D Classic with the newer bodies with smaller sensors and greater pixel densities, the larger sensor with the larger pixel wells wins for cleanness and all that entails.

You assume that it's the larger pixel wells that causes the 5D classic to win, but that's incorrect. The reason it "wins" is entirely and solely because of the larger sensor size. The pixel size has nothing to do with it. If you take away the sensor size factor by cropping the 5D (but still leaving the pixel sizes the same), you'll find that it loses badly compared to "greater pixel desnities".

This has been demonstrated before, even comparing the 5D against 20D. Here's a few examples:

20D (1.6x) vs 5D (FF) noise equivalency

S3 IS (6x) vs 5D (FF) noise equivalency

30D @ 85mm vs 5D @ 135mm vignetting / edge sharpness / noise equivalency

400D vs FZ50

40D vs 50D

The reason why so many people are misled into thinking pixel density causes noise is that they perform unequal comparisons. The biggest mistake is to use 100% crop, which is comparing two images at completely different magnifications (spatial frequencies), like printing one at a larger size then another, then cropping for comparison. The comparison should be at equal spatial frequencies. Other mistakes include unequal raw conversion, processing, sensor size, technology level, etc.

The 5D2 has 2.56X more area than the 50D, so it has much less noise and thinner DOF when used at the same f-number. If you take away the sensor size advantage by cropping to the 50D dimensions, then all it is left with is the pixel size difference, which is no advantage at all.

If you can afford the 5D2, it's great (I love mine).

This might interest you...

http://jasonhollister.​com/blog/?p=44

Well, hmmm, I see it both ways, and have seen it both ways (shooting with the 30D, the 1D3, and the 5D. It's both enlargment advantages and pixel-level signal-to-noise advantages.

Look at it this way: if I take a shot with my 30D at ISO 1600 and a shot with my 5D at 1600 with the same exposure, I could compare in this way: print the 30D shot at 12"x18" and the 5D shot at 20"x30". Hang them together, and look at them at the same viewing distance, close enough to see detail, and the 5D image will be cleaner. To me, that's a meaningful comparison.

What about the "Black Dot" problem with the 5d Mark II? Has this been solved, or was it never a real problem to begin with ?

J1000 wrote in post #7751153
What about the "Black Dot" problem with the 5d Mark II? Has this been solved, or was it never a real problem to begin with ?

They released firmwar some time ago that addressed this. I don't know the technical details, but ever since it came out the complaints went away, so I assume the fix worked fine.

tonylong wrote in post #7751212
They released firmwar some time ago that addressed this. I don't know the technical details, but ever since it came out the complaints went away, so I assume the fix worked fine.

That is what I have been thinking, but I also saw some references that were critical of the fact that Canon said it was "mitigated" rather than fixed (e.g. put a color other than black into the black spot, so that it is less noticeable)

I should mention one more aspect of having the pixel level high signal-no-noise ratio advantage that you see with full frame images: something that I refer to as "pixel richness". Many full frame users comment on the fact that their images straight from the camera have more "pop", or "clarity" or "richness" than they were used to with the crop camers, and that consequently led to less post processing.

A lot of people question this, suggesting it's psychological, and more point out that you can even things out with post processing, but really there is logic to this perception. Each pixel has more light gathered leading to that higher signal to noise ratio, and meaning that the "raw material" is purer and richer than that gathered by the more compressed sensor. What would the expected outcome be? Well, to me it would be expactly what people report.

That doesn't mean that it will always be noticeable, but over time I do notice the difference between my 5D and 1DM3 files (low pixel density) and my 30D files. I've heard the same from people who have been 40D files.

No one is suggesting that the 30D/40D/50D cameras are bad, but merely that the images from the bigger sensor do have an inherent quality boost.

J1000 wrote in post #7751231
That is what I have been thinking, but I also saw some references that were critical of the fact that Canon said it was "mitigated" rather than fixed (e.g. put a color other than black into the black spot, so that it is less noticeable)

Yeah, I'm not sure, 'cause like I said I don't know the details. Of course, all pixel values in digital sensors are interpolated (that is the ones that use the Bayer sensors) based on neighboring RGB values, so the black dots were likely an interpolation glitch and was relatively easy to fix.

J1000 wrote in post #7751231
I also saw some references that were critical of the fact that Canon said it was "mitigated" rather than fixed (e.g. put a color other than black into the black spot, so that it is less noticeable)

That critique is justified to a certain degree.

The underlying problem of that phenomenon appears to be bleeding: the charge of extremly overexposed pixels can not be contained (literally) in the well and it "leaks" out, and reaches the neighbouring pixels. The pixel values start normally around 950, but those black pixels were much-much lower, starting from zero. I don't know what exactly happens, i.e. how the bleeding caused those low pixel values. The affected pixels are always to the right and downwards from the oversaturated pixels.

Anyway, it is true, that the underlying problem has not been solved, probably it can not be solved, it is inherent to the particular sensor design. On the other hand, are pixels adjacent to the extramely overexposed pixels really so important? It IS important, that they don't stick out, i.e. not be black beside the "shiny" pixels, but otherwise, if they blend in their surrounding, then one should not complain - the overexposed pixels are incorrect anyway, and nothing can be done about that.