Why 1.6 Crop?

[Daniel Browning]

It sounds like you generally agree with Mark and Jon, but you just want to emphasize the importance of technology and sensor characteristics. If so, then I agree, but I want to emphasize the value of knowing what happens when the other factors are equal.

The reasons used to support the 'less light with smaller sensor' above are all dependent on the characteristics (not just size) of the sensor itself, not the amount light falling on it.

You're definitely correct that it's a combination of multiple factors. Generally, the single most important factor is the total amount of light falling on the sensor. Characteristics of the sensor are definitely important, and in some conditions they are enough to outweigh sensor size. (For example, the 5D classic beats the 7D in ample light, but the 7D wins in low light.)


Anyway, that doesn't take away from the importance of understanding what happens when the performance per area is the same and you vary the area.

Yes, croppingmakes pictures appear more noisy - it doesn't make then noisier, the noise is there (digitally), but enlarging them makes it more visible in the image.

I don't think that is a valid distinction because the same is true about sensor size. The enlargement ratio between a small sensor and a large sensor is the same as between a small digital crop and no cropping (if no correspond to the same physical sensor size and output display size).

Smaller sensors have more noise, but that depends on the 'all other things being equal' clause

True. For example, the a 1.6X crop from the 5D2 has about the exact same amount of light and noise as the 7D. So the AOTBE clause holds true in that case.

smaller pixels have more noise, yes, because their photon depth is shallower (lower eV) meaning they have a lower signal to noise ratio. But, if you pack 4 'shallow' pixels into the same physical area as one
'deep' pixel, you'll capture nearly the same amount amount of light
(some is lost due to inter-pixel spacing), with a tradeoff of more
noise and less dynamic range.

Well, this is a side issue, but there is much evidence that this is only partly true. Low gain read noise and dynamic range are invariably better with smaller pixels. It's only in low light where there is probably a correlation between pixel size and read noise (but even then there are a lot of times when that is not the case). See the following thread:


http://community.the-digital-picture.com/forums/t/1055.aspx

you'd have to apply a correction like that when using the same lens on an old 1Ds MkI vs. the forthcoming iDs MkIV, or a 50D vs. a Rebel XT, etc., as the equal-sized sensors in each pair have very different light-gathering ability.

Actually, there has been very little difference in light gathering ability for several years now: less than 10%. The 50D does about 30% better than the XT, but the 50D and 7D are within a few percent of eachother. That means noise in midtones at low ISO are not going to improve at all. It's dynamic range and low light noise that will get the improvements.


Please take a look at the following comparison I made:


5D2 crop comparisons


In this comparison, the performance per area is the same (i.e., read noise and QE per spatial frequency in the output image from a given crop size).


Here's the Full Frame at 111mm f/6.3 ISO 1600:





And here's the 1.6X at 70mm f/4.0 ISO 640:





They are very similar, despite the fact that one is ISO 640 and the other is ISO 1600. You would get the same results from any test where performance per area is the same, like the 5D2 and 7D, 1D3 and 40D.

[Keith B]

Yes, but Daniel do you know why Canon chose 1.6 instead of say 1.5?

[Daniel Browning]

Yes, but Daniel do you know why Canon chose 1.6 instead of say 1.5?
<div style="clear: both;"]</div>

I don't know. It may have been a difference between the semiconductor stepper technology built by Canon and Nikon (Nikon supplies the steppers to Sony and others, but Canon builds their own). Perhaps the largest reticle at the time they came up with the D30 was only 22.7mm. Or perhaps it was because Canon wanted to leave more room for options around the outside of the sensor frame, such as a higher number of masked pixels (which are useful for reducing some types of noise), though I'm not aware that they are actually doing that in different proportion to Nikon. Or perhaps it was just a slightly different balance between cost and performance. 14% lower cost and 14% lower performance. Maybe it was something else entirely, I don't know.

[Keith B]

Fair enough. Thanks.

[neuroanatomist]

Thanks for the detailed and cogent explanations, Daniel! Much appreciated.

[Colin]

I would have preferred a 1.4x crop.... Easier to think of in terms of stops...


But that would have been too close to the 1.3x crop, and we can't have THAT! :P