Format size and spatial resolution: Replies to Jack Hogan

Started 9 months ago | Discussions thread
ForumParentFirstPreviousNextNext unread
Flat view
Anders W
Forum ProPosts: 17,050Gear list
Like?
Format size and spatial resolution: Replies to Jack Hogan
9 months ago

The thread in which Jack posted this and this reply to me expired. So I answer here instead.

Jack Hogan wrote:

Nor am I. I think the best one can do when attempting to compare spatial resolution from different formats in general, as opposed to specific instances, is to compare data collected with the same methodology, across cameras of roughly the same generation with roughly the same AA filter strength. It's very difficult to do with information freely available on the web, so one is tempted to cherry pick.

We're in full agreement here.

What makes you think that this comparison is more apples to apples than mine

That 's a good question. In the following chart I tried to include tests done with just a few cameras before around 2011, when the trend to weakened AA's started - a trend that as I mentioned distorts the comparative relevance of MTF50 readings substantially, and it will continue to do so untill they are all AA-less. I specifically chose an inexpensive FF prime as a reference, a lens that has been around forever, as a representative of what photographers have been looking for in a lens for 100 years - so the Nikkor 50mm/1.8D is a decent lens, but no Sigma 35/1.4. Its performance, as measaured on APSC, was plotted in mFT terms as the target spatial resolution to achieve (the red line). Then I plotted every single prime I could find from mFT cameras of roughly the same generation obtained through the same methodology, in order to get the law of averages working.

I agree that this is a reasonable choice based on the premises you mention. But current MFT lenses show significantly higher MTF values on the 12 MP E-PL1 sensor than the 50 lp/mm reached on earlier FT/MFT sensors. As you know, Lenstip now reports peak values of over 80 lp/mm for some MFT lenses on the E-PL1 whereas I haven't seen any indication that things have changed to nearly the same extent on the FF or APS-C side of things. Exactly why that is is of course hard to tell. Could have been that earlier FT/MFT sensor had an unusually strong AA or something like that.

and what size penalty are you talking about? Yes, a lens made for MFT has to resolve twice as well per mm on the sensor as a lens made for FF. But it also has to cover an image circle with only half the diameter.

Yes. However the admittedly better and smaller glass does not appear to be able to make up for the sensor size disadvantage in most situations. That means a much higher requirement in linear spatial resolution that often can only be obtained from a fewer number of pixels to boot. When I was able to compare apples to apples the combination of these two factors appeared to typically be an insurmountable hurdle to jump for the smaller format - of course there are exceptions but I was interested in a generic evaluation, not the exceptions to it. Here is some data from raw captures off DPR's own new scene - not a rigorous exercise with any different variables to be taken with all the provisos of the case, but at least the methodology should be the same:

Yes, but that comparison is of course affected by the pixel count of the various sensors, which is typically higher for larger sensors. In order to say anything about the impact of sensor size per se, we should compare at the same pixel count and, to the extent possible, same strength of AA filter. And when we do, there isn't much of a difference even in this case as far as I can see.

So it appears that when talking generically about different formats of roughly the same generation the larger format typically wins out in terms of spatial resolution when the final image is viewed at the same size. And it wins big if it is not constrained by the smaller format's limitations for the sake of equivalence, as one can see in the EM5 vs D800 readings for instance.

If FF sensors win out here, it's largely because the sensor resolution is higher (36 MP versus 16 MP in the case of the D800 versus the E-M5, for example). And the resolution advantage is in this case bought at the expense of higher noise. Note that the figures you get from DxO (unless you press the "screen tab") are for a resolution normalized to 8 MP, not for 36 MP versus 16 MP.

And what about the following comparison between the Panasonic 20/1.7 on an E-M5 (16 MP, AA filter) versus a Summilux 50/1.4 (the very best of all the FF normal lenses tested by LensRentals in the test I link to below) on an M9 (18 MP, no AA filter). The figures are lp/ih at MTF-50, center/average, based on unsharpened output from RAW files. Here, the cheap Panasonic runs neck to neck with the far more expensive Leica in spite of a slight disadvantage on the sensor side (AA filter plus slightly lower resolution).

Panasonic 20/1.7 on E-M5 (16 MP)
1.7 870/735
2.8 1050/875
4.0 1075/880

Leica Summilux 50/1.4 on M9 (18 MP, no AA filter)
1.4 600/530
2.0 950/740
2.8 1025/860
4.0 1110/980

Sources:

http://www.lensrentals.com/blog/2012/01/the-great-50mm-shootout

http://www.lensrentals.com/blog/2012/05/wide-angle-micro-43-imatest-results

I wasn't aware that the M9 is AAless but yes, that 20/1.7 is good value for the money. I can't explain lensrental's results even because the Summilux figures are more what one would expect from a Nikkor 50mm cheaper than the panny.

Not sure I follow you here. The figures from LensRentals are generally well in line with those of Lenstip even in absolute terms. Here's an example using the Nikkor 50/1.4G on a D3X.

Lenstip has center/edge lp/mm figures of 24/19 at f/1.4 and 41/31 at f/4. LensRentals has center/average lp/ih figures of 560/500 at f/1.4 and 900/830 at f/4. Divide these figures by 24 and you have 23/21 and 38/35. So everything is pretty well in line with what you'd expect.

I looked (for about 10min) but I could not find how they prepare the raw files for Imatest. Some, like Photozone, demosaic and sharpen them. Others use in-camera white balance. Others report only horizontal or vertical results. etc. etc*. Any of these distort individual resuts substantially and could explain some of the expected difference.

No it couldn't. In this regard, the methods used by LensRentals are essentially the same as those of Lenstip: Unsharpened output based on RAW files. See for example under "Methods" in the first of the two links to LensRental tests I already gave you.

Or perhaps the Panasonic is just an outstanding little lens But as I mentioned my comments are meant to be read generically about the relative performance of the formats, not about specific exceptions.

No. That explanation doesn't work either. The 20/1.7 is certainly among the best MFT lenses MTF-wise. But I pitted it against the best FF lens, not the average. If we take the Nikkor 50/1.4G as an example of the average, the Pany beats it as you can see, in spite of the Nikkor being tested on a sensor with higher pixel count (24 MP versus 16). Furthermore, quite a few of the MFT primes show values similar to that of the 20/1.7 (in most cases in the testing by Lenstip as well as LensRentals).

More than happy to revise my findings if data supporting different conclusions surfaces. But for now at least, if one is fixated on spatial resolution it appears to me that the larger the format, the better.

Well I think these data have already surfaced. Let me know if you want more examples, and I'll happy to do what I can in order to provide them.

This is an interesting question I was pondering and I think it would depend on methodology and comparability of the data. I took the ratio of the D200 and E3 effective pixel pitches, but the rationale was a little more convoluted than that.

Lenstip figures are averages of vertical and horizontal MTF50 readings. 'Older' cameras with relatively stronger AAs tended to produce fairly evenly matched perpendicular MTF50 radings. For instance, using my data, the D800 read 0.223346 cycles/px roughly horizontally and 0.233469 cycles/px vertically, and the EM5 0.193970 and 0.209135 resp. But newer cameras with weaker AA filters sometimes show quite large differences between the two. For instance the EM1 0.243925 and 0.215788 resp. So the EM5 in that setup would show about 52.0x55.9lp/mm (HxV) vs the EM1's 65.0x57.4.

Lenstip would average these figures. So for the chart based on Lenstip data in the earlier post, it seemed to me that we would want to average the H and V sensor ratios in order to present consistent figures. I used sensor data for the D200 (16x24mm) and E3 (13x17mm) from DxO - because those were the cameras mostly used - which average out to about 1.32x. But perhaps for cameras of roughly the same number of MP it's more intuitive to think about the ratio of effective pixel pitches (6.13/4.61 microns), which unsurprisingly gives the same result.

Well, we want to compare at the same final display size (same area covered by the final photo), right? Going by the pixel pitch as you do, accomplishes this only if the pixel count of the two sensors is exactly the same. Going by the square root of the ratio of the one area to the other area accomplishes it even if there are minor differences in pixel count.

 Anders W's gear list:Anders W's gear list
Panasonic Lumix DMC-G1 Olympus OM-D E-M5 Olympus E-M1 Panasonic Lumix G Vario 14-45mm F3.5-5.6 ASPH OIS Panasonic Lumix G Vario 7-14mm F4 ASPH +21 more
ForumParentFirstPreviousNextNext unread
Flat view
Post (hide subjects)Posted by
ForumParentFirstPreviousNextNext unread
Keyboard shortcuts:
FForum PPrevious NNext WNext unread UUpvote SSubscribe RReply QQuote BBookmark post MMy threads
Color scheme? Blue / Yellow