Imatest interpreting and lense testing

The various specs are so confusing. Imatest which pcmag.com and imaging-resources.com uses and dxomark which dxo and dpreview now uses are different metrics for the same thing. How good is a particular body and lense combination. Truth is that a great body (NEX-7) can turn in a low resolution camera with a mediocre lenses (it’s kit lense) produces an effective 9-10 megapixels whereas a very good lense (Olympus prime) and body (MD5) can produce a 12 megapixel image.
What is more complicated is that resolution changes dramatically from the center (where it it shapest) to the edges (worse) and the corners (even worse). And the focal length on a zoom and the aperture have a dramatic effect (on a Sony RX-100, F/4 is 30% sharper than wide open at F/1.8).
The most confusing part is that the measures are different so Imatest likes to talk about lp/ih that is line pairs per image height and DXOmark likes to talk about effective pixels (without defining what that means), although the idea of having a single figure of merit is sure useful when you realize that resolution varies substantially across lenses, apertures and focal lengths!
And the other problem is whether you are using JPG or RAW and what image sharpening you are adding. As my buddy Scott points out, you can increase resolution from 2400 to 3500 just by sharpening! Also, the new DxO tests are confusing as well it seems that this is mainly because sharpening is such a big factor. Specifically it looks like the d800 dxomark MP ratings are low (they are undersharpening), the 5d3 are oversharpened. Looks like photozone.de has the best measurement. Maybe the large lesson is that sharpening is a huge factor in effective resolution. You need sharpening by the way because the anti alias filter intentionally blurs things. Nowadays the more modern cameras are getting rid fo the antialiasing and instead removing moire in software in those cases where it exists. Way smarter as antialias really reduces resolution.
So how is anyone supposed to think about this? Well, perhaps the first thing is to get some sort of simplification across Imatest and DXOMark particularly in those cases where the Imatest thing has me starting at acutance charts looking at resolution maps across an image. Ugh.
First question is how to think about Imatest in terms of effective resolution. So what is the actual line height for a given sensor. You can look at this as a pretty easy rough math problem (because hunting for the exact resolution vs image height spec is actually pretty hard, but here are the rough numbers)
image height = sqrt( megapixels / (aspect ratio))
megapixels = height^2 * aspect ratio
In most modern cameras, they still use the 35mm film aspect ration (24x36mm or 4:3 aspect)

MP LH
8 sqrt(8*3/4)=sqrt(6) 2,449 lines
12 sqrt(12*3/4)=sqrt(9) 3,000 lines
16 sqrt(16*3/4)=sqrt(12)=2*sqrt(3) 3,464 l/ih
20 sqrt(20*3/4)=sqrt(15) 3,872 l/ih
24 sqrt(24*3/4)=sqrt(18)=3*sqrt(2) 4,242 l/ih
36 sqrt(36*3/4)=sqrt(27) 5,196 l/ih

What does this all mean, well if you look at Imatest results from PC Magazine, the resolutions from their various reviews and using the formulas above makes it pretty clear that the sensor size is really not super relavent for these ILCs and point and shoots. What matters is the glass, although these tests are really had to interpret as the writing is vague and they are using JPG, so it is not clear what sharpening they are using. Scott reports that for instance before sharpening (which really corrects what an anti-aliasing filter removes) that you can double resolution!
Nonetheless, assuming these are out of the box JPG settings:
Sony RX100. 28mm. F/1.8, 2126li/ph center that is the worse case of wide open but at center (you can invert the formula above to get an effective point megapixel resolution. (2126^2 * 4/3 = 6MP equivalent) against a 20MP lense. That is pretty incredible for a wide open lenses. The sweet spot for this lense is around f/2.8 which is at 2,300 lines = 7MP). It is nice to do this as a ratio of the potential resolution possible which is 3800 lines in the RX-100 case. Now that’s a best case since it is in the center obviously.
Ricoh GR. Interesting to see that at wide open the 28mm fixed lense at f/2.8 resolves about teh same at 2105 li/ph (about 6MP) so not much different than the RX100. It doesn’t have an anti-aliasing filter like the Coolpix A (the modern trend as it improves resolution). The edge score stays at 1,927 li/ph (5MP equivalent). They have a wide-angle adapter to get to 21mm lens, without significant light loss which is 2,016 li/ph at 21mm lense effectively.
Samsung NX300 with 45mm f/1.8 2D/3D. Interesting to see how on Imatest, there is sort of limit of 2000 to 2400 lines (5-8MP) for this class of camera. AT f/1.8 in the center it is 2,205 with a maximum resolution of 2,720 lines at f/5.6 (10MP) for a 20MP sensor. These are optimistic as these are center only measurements of course.
Samsung NX300 with 18-55mm f/3.5-5.6. He says he is using a center weighted test so maybe this means not center only. In any case, wide open at 18mm f/3.5 of 2095 li/ih (5.8MP vs 20MP sensor) with the edges at 1,536. The best wide is at f/8 giving a 2360 lph average (6.8MP) with 1900 lph at the edges. The lense gets worse as you zoom in (which I haven’t seen before and then gets better all the way racked out). So at 35mm f/4.5 wide open, it is 1857 lph (4.6MP) center averaged and 1,658 lph at edge improving to 55mm with 2,241 at f/8 (6.7MP) with 2,000 at the edges. Noise is also probably a stop worse than an NEX-6 although, he doesn’t give specific analytic metrics.
Samsung NX210 with 12-24mm f/4-5.6. This is a wide angle but slower lenses. It was tested on an older Samsung which also had a 20MP sensor. At 12mm f/4 wide open, it got (not clear but these look like center only measurements) of 2,385 li/lh
Photozone.de seems to have at least a more uniform test:

Share
%d bloggers like this: