Limited by light (or problems with red)

[Fine_Art]

Most of us these days are diffraction limited at the sensor. FF sensors are typically 24MP or higher, APS-C sensors are 16MP or higher. This means our pixels are 6 microns or smaller. Looking at the attachment, assuming I did the maths right, we see that to fit a spot of a color into the 6 micron pixel we need to be at f2.8 to include red, f4 if we can live with red fringing edited in software, f5.6 to approximate everything  but violet with software, etc. Unless you are using the best glass wide open with possibly focus stacking, the software is doing an amazing amount of heavy lifting interpreting colors that are painted wide jiffy marker style over the pixels. No wonder our images look far more realistic binned with a 50% down-sample.

The general implications are:
Expect red fringing unless your lens is diffraction limited at <f4, APO or not.
Lenses that are optimal at f8 (anything cheap 5.6 or more wide open) mean you will be binning your shots if you are particular about detail.
tethering with focus stacking software is becoming important.

or give up and try to deconvolve the problems which may mean running separate deconvolution radiuses for R,G,B.
or give up and down-sample using stitching to get a big print.

or ignore the fact we have engineering marvels at the limits of optics focusing on making images that will be forever small.

What have I missed?

[Fine_Art]

I should add that the perceived higher image quality of medium format is probably more due to the light falling into the larger pixels correctly more than anything else.

[Bart_van_der_Wolf]

What have I missed?

Hi Arthur,

Allow me to add a few additional thoughts.

1. Red wavelengths do produce larger diffraction patterns than shorter wavelengths, but they only amount for some 21% of the luminance weighting (green=72%, blue=7%). A more meaningful impact on (luminance) image resolution can be expected when one calculates with something like 555 nanometers. I often calculate with 564nm as a 7%/72%/21% weighted average between blue, green, and red, to allow a slightly larger influence of red (because of foliage and such subjects that are a bit more yellow than 555nm green).

2. The intensity distribution within the Airy pattern is not uniform, but rather peaked at the center (even more than a Gaussian distribution).

3. Due to the Bayer CFA, color is more point sampled than area sampled, which will boost the MTF response (but also cause aliasing). Demosaicing does level the resolution differences between the R/G/B channels, because luminance reconstruction is very good and that resolution is reintroduced in the color interpolation.

4. The optical low-pass filter (OLPF), if present, will add to the diffraction pattern blur. Residual lens aberrations also add blur.

In practice, I do not see much impact until the (564nm) diffraction pattern diameter (first zero of the Airy pattern) exceeds 1.5x the sensel pitch. At that point, enough energy is distributed to neighboring sensels to reduce micro-contrast beyond what is recoverable. That is therefore only destructive for higher spatial frequencies (=lower contrast), lower spatial frequencies will lose contrast but may be recoverable by deconvolution sharpening. Low noise image captures will offer the best restoration opportunities.

Cheers,
Bart

[Fine_Art]

Thanks Bart, I was hoping you would fill in a few blanks, as well as others.

One area I have noticed problems is in red flowers. DSLRs really struggle to get detail. Often the parts with detail are the veins which are not quite red. I have decided to always focus stack these subjects at f2.8-f4.

[Bart_van_der_Wolf]

Thanks Bart, I was hoping you would fill in a few blanks, as well as others.

One area I have noticed problems is in red flowers. DSLRs really struggle to get detail. Often the parts with detail are the veins which are not quite red. I have decided to always focus stack these subjects at f2.8-f4.

A big issue with Red flowers is the risk of oversaturation of the Red channel, and/or underexposure of the Blue channel. Do check for that as well, because it may result in gamut clipping on lower gamut output modalities. The Blue and Red channels are often underexposed in Raw, but with White-balancing especially Red can oversaturate very easily. Add some diffraction which is worst in pure red, and you'll have a nice set of potential trouble.

Indeed, focus-stacking at somewhat wider apertures (based on, say, 1.5x sensel-pitch diameter for the Red diffraction pattern diameter) can help. Do remember that with close-ups/macro, the diffraction pattern size is also magnified.

A plugin such as Topaz Detail, allows to selectively brush in (or out) some (deconvolution) Deblur and/or Saturation which allows to target some of the trouble areas.

Cheers,
Bart