Chuck Norcutt wrote:
> What got me on this point to begin with was DPreview's test of the Canon
> 70-200/2.8L ... So, if this lens can just barely handle a 5D it certainly
> can't handle the pixel density of a 50D (15 MP on an APS-C size sensor). At
> f/5.6 to f/8 the 5D Mk II will pull a bit more out of the lens than the 5D
> but not a great deal. To utilize the resolving power of the 5D Mk II you
> need very good primes. ...
I'm not sure I agree with your conclusions. I suspect that the following
considerations aren't meaningful for the vast majority of actual image
making and display. However, for those few working at large
magnifications and/or the larger number simply afflicted with terminal
technophilia or untreated addiction to pixel peeping, they may be, or at
least seem, significant.
First, I have issues with coming to such broad conclusions considering
the methodology used.
1. Judgments/measurements are made with a camera with AA filter in
place. That means one could adapt some industrial super lens capable of
resolving much more than the sensor, but the system resolution they
measure, while better than regular lenses, would be nowhere near the
lens's true ability. As long as there is an intentional bandwidth
limiter in the system, no fully meaningful conclusions may be drawn
about individual parts.
2. The lppmm standard, while a good method to compare cameras with
different sensor sizes, is not ideal for lens comparisons or judgments
about lens vs. sensor resolution.
3. A nit pick. How can the system out resolve the "theoretical
resolution of the sensor", Nyquist frequency? If theory doesn't agree
with practice, why even mention it? My real complaint though is that it
gets the reader thinking about absolute limits, in a test environment
where they don't exist.
There are also a few other considerations that make coming to such
conclusions from these test results questionable to me:
1. Back in the area of #1 above, system bandwidth is a complex function
of the bandwidths of the components. It's not like hooking a garden hose
onto the end of a fire hose, where throughput is strictly a function of
pressure and the physical characteristics of the garden hose. Even the
formula someone posted a little bit ago is probably too simple in the
real world. In addition to which, we don't have the numbers for
individual components to plug into it.
Increase the resolving power of any one component and system resolution
goes up. If the improved component was the poorest, it goes up a lot,
but not as much as the % increase in the component. If it was already
the broadest component, bandwidth increases only a little. But it does
change.
2. These luminance only, test pattern tests don't say a thing about
spatial color resolution. If Foveon sensors had fulfilled their apparent
potential, this wouldn't be an issue. However, all but a tiny proportion
of digital cameras use a Bayer array. So depending on color and
directional orientation, the color resolution may be quite a bit less
than B&W. Demosaicing algorithms are quite good, but the fact is that
only one of the color channels for any pixel is a actual measurement.
The rest are interpolated guesses.
Since there are a few Foveon sensor cameras about, it's possible to make
actual comparisons. <http://www.ddisoftware.com/sd14-5d/> The SD14 vs.
5D results convince me that larger Bayer array sensors will have better
color IQ than smaller ones, regardless of luminance test target tests.
As the lenses we are working with project the magnificence of their
images on all sensor sites equally, an improvement in sensor color
resolution will show up in visible image quality.
3. There is an unspoken assumption in most web discussions of color
accuracy in digicams, but it is wrong. When we talk about sensors, we
act like the array filters are at least close to perfect. It just ain't
so. Look at the spectral responses of even the best quality photographic
filters. They aren't sharp cut-off tools, not even close. How good do
you think the gazillions of tiny little, mass deposited Bayer array
filter in our cameras are?
So the fact is that wavelengths of light falling on sensors are only
majority reported by the appropriate sensor sites. There are also
minority values misreported by adjacent sensors for other colors, with
the amount of error depending on how far the actual color is from the
center of the intended sensor for it.
I don't know how big these errors are, but suspect that they are
significant in an absolute sense. We tend to be unaware of them because
we have been viewing the same sort of errors on color film and because
nobody is publishing about them. In any case, more, smaller sensors will
tend to reduce these errors.
4. No matter what the resolution chart tests say, all digital capture
involves sampling and that sampling always reduces
sharpness/contrast/clarity/whatever along any edge that is not exactly
aligned with the sensor array. If you work out the angular requirements
for that on a large sensor, you will see that it applies to essentially
all edges. As I've written about a while ago on Zone-10, that means all
images captured digitally have poor edge contrast relative to the
nominal resolution. {Starting with pp6 here.
<http://zone-10.com/cmsm/index.php?option=com_content&task=view&id=149&Itemid=1&limit=1&limitstart=1>}
Sharpening generally can do a wonderful job of improving apparent edge
sharpness and visible resolution. However - for exactly the same reasons
the digital image has edge clarity problems in the first place,
sharpening has to work on adjacent pixels as well. So while an
improvement, sharpening is not a true replacement for simply increasing
the sampling frequency, i.e. more pixels.
When you add the issues in #2 & 3, sharpening will usually be working
with significant numbers of pixels with already compromised color data -
and changing it still further.
In conclusion, I suspect that, whatever the test targets seem to
indicate about resolution of simple B&W bars, more sensor sites will
continue to improve overall IQ in other images areas, including color
resolution.
I'm not suggesting that such tests are not quite useful. I think the
simple tests are pretty good analogs of overall lens performance and a
good practical way to compare them. That they don't say it all is
testified to by the decades of testimony of practical photographers
about meaningful difference between lenses that aren't reflected in
simple resolution tests.
I am suggesting that a simple conclusion about whether a give sensor
"out resolves" a given lens, and therefore whether there will be any
benefit from a higher resolution sensor, based on these tests, is not
supported by the methodology.
Moose
--
_________________________________________________________________
Options: http://lists.thomasclausen.net/mailman/listinfo/olympus
Archives: http://lists.thomasclausen.net/mailman/private/olympus/
Themed Olympus Photo Exhibition: http://www.tope.nl/
|