- Sense or Size -
... which matters most ?
The battle of the sensor sizes
I often meet people who are anxious about the fact that they are using a so-called cropped sensor. When I tell them that for most applications there is nothing to worry about, as the results will not be noticeably different, they mostly refer to iso charts proving how much better large sensors are. Full Frame being the magic word. When I try to explain them that with smaller sensors you often can use lower iso's, and thus obtain similar noise levels, the typical reaction is one of reject and disbelief. So I thought it might be a good idea to demonstrate this point via a simple test. The goal of this test is actually a bit more general : to understand what matters most in the final image : the mere sensor size, or the lens.
We will set up the test the way they should be done, that is by using so-called equivalent settings. Here's an example of theoretically equivalent settings :
Full Frame @ iso 400, 70mm f/2.8
1.4x crop @ iso 200, 50mm f/2
2x crop @ iso 100, 35mm f/1.4
Now I don't have a 1.4x or 2x crop camera, but I can emulate a 1.4x crop factor by comparing a slightly cropped full frame with a 1.6x camera, both from the same brand. Slightly cropping the frame also matched the pixel counts, thus further ruling out sensor differences (other than size). Finally, I bumped up the iso in order to make some noise appear. So this will be the practical test setup :
(1.14x cropped) Canon 5d @ iso 1600 + Tamron 28-75mm f/2.8 @ 70mm f/2.8
- versus -
Canon 40d (1.6x crop) @ iso 800 + Canon 50mm f/1.8 @ 50mm f/2.0
I took exactly one shot with each camera. These are the results :
Tamron 28-75mm f/2.8 @ 70mm f/2.8
(cropped) Canon 5d @ iso 1600
Canon 50mm f/1.8 @ 50mm f/2.0
Canon 40d @ iso 800
As one can see in the results above, using equivalent settings gives images where field of view and depth of field match up nicely. So what about the image quality ? Detail and noise are very close. Perceivable differences are due to lens colour rendition (the results look the same but the white balance settings are actually very different), lens distortion, lens bokeh (out of focus smoothness) and lens focussing accuracy.
If pixel counts are roughly the same, the mere sensor size does not matter all that much as far as overall image quality (including noise) is concerned. However, the lens has to be sharper for a cropped camera to provide the same information to the sensor (as the pixels of the sensor are closer together). The lens also has to have a smaller f-number in case one wants to be able to achieve the same shallow depth of field. And the viewfinder of the cropped camera needs a higher magnification to give the same user experience. So as sensors get smaller it typically becomes harder to achieve the same result (exception : in body image stabilisation is easier to achieve for smaller sensors). But it can be done, at least to a certain extent, as has been proven by manufacturers like Olympus and Panasonic using a 2x crop system.
For the interested, the following details show how I used a full frame camera (Canon 5d) and a 1.6x crop camera (Canon 40d) and available lenses (Canon 50mm/1.8 and Tamron 28-75/2.8) to set up a test case for a crop factor of 1.4x (instead of for the more obvious 1.6x). In this way, it was more easy to select matching f-stops and iso settings, and to level the pixel counts. The trick was to slightly crop the 5d image. One thing I did not expect is that the Tamron when set at 75mm actually behaved like a 70mm lens (at least when focussed on items some 2-3 meters away). So when I wrote 70mm above, I actually had the lens set on the 75mm mark. One can start arguing over this but in the end, what counts, is that with two single shots, one on each camera, I managed to achieve a really close match, close enough to prove the point as decribed in the intro.