Re: Modern Sensors Missing half the AA: in the Horizontal or the Vertical direction?
Jack Hogan wrote:
fvdbergh2501 wrote:
This extracted only the green photosites, completely ignoring red and blue, so wavelength-depended diffraction is avoided. (I did check the red and blue channels too, and focus was definitely optimized for green).
Hmm. Are you saying that the differences in the various channels are mainly due to diffraction effects? In this case by measuring MTF50 of the 3 channels individually and comparing them one might be able to calculate/extract the impact of diffraction vs aberrations. Interesting.
And what effect does only measuring the green channel have on MTF50 readings - since we are effectively cutting the sampling rate in half?
Well, there are many possible causes for observing different MTF curves in different wavelengths. Diffraction is always present; I measure 53.63 lp/mm in green, and 55.65 lp/mm in blue if I generate synthetic D610 images. The single largest factor is probably the lens design (achromatic, apochromatic): different wavelengths focus at difference distances from the focal plane (the sensor). The hope is that you can bring all the wavelengths close enough to perfect focus so that no one notices, but even an apochromatic design can one really have three wavelengths in focus, with varying degrees of defocus on all the wavelengths in between.
Of course, the expected increased resolution at blue wavelengths (lower diffraction) can be countered by focus (i.e., green is more in focus than blue), which appears to be the case with these D610 images. I do have actual D7000 photos where the blue channel produces slightly higher resolution than the green.
In terms of noise, yes, a using only the green photosites cuts the number of samples in half. That would put the minimum edge length at about 60 pixels (with 100 pixels being more comfortable), but the DPR charts have >600 pixels along the edge, so in this case it works well even for the blue or red photosites.
So here is the trade-off: we know how noise affects MTF measurements, i.e., the mean value over many individual measurements is unbiased, but the deviation from the mean can be quite large for any single measurement (at high noise levels). A difference between the blue and green (for example) focal plane position relative to the sensor would be systematic, i.e., it would not decrease with repeated measurements. This means that an un-demosaiced image (dcraw -D) followed by white balancing will produce an edge that is a blend of three individual edges (red, green and blue). White balancing, in itself, does not affect edge sharpness, so a perfectly white-balanced mosaiced edge image will probably produce a weighted MTF curve (25% red curve, 50% green curve, 25% blue curve).
If absolute accuracy is important, and multiple images are available, then repeated measurements followed by single-channel analysis (e.g., mtf mapper's "--bayer green" option) would be the best strategy.
If multiple images are not available, then edge length will dictate your choice: if the edge is sufficiently long (e.g., 200 pixels for green-only analysis), use single-channel analysis, otherwise use the white-balanced analysis.
The worst possible method is, of course, to use a demosaiced image, because nobody knows exactly how the MTF curve could be distorted by some non-linear interpolation scheme that the demosaicing algorithm might employ.
Incidentally, the overall resolution is somewhat low. Simulated >>results at f/5.6 indicates that we could see values of up to 62 >>lp/mm with an OLPF, and about 86.3 lp/mm without an OLPF.. The >>62 lp/mm is consistent with what I saw on real photos captured >>with a D7000 (4.73 micron photosite pitch, vs ~5.09 micron >>pitch on the D610).
The D610's pixel pitch is about 5.9 microns. Plus the 85mm:1.8G >is a good lens, not an outstanding one. And there is always the >possibility of operator error (say incorrect focus peaking? >That's clearly the case in the D5300 captures for instance). >Although results are fairly consistent with the similar-sensored >A7 and RX1R (different lenses though).
Whoops! Good catch. I see that I used 5.9 in generating the simulated images, but 5.09 when reporting some of the lp/mm ratings. So the simulated results should be 53.63 lp/mm with an OLPF, and 74.5 lp/mm without an OLPF. The 44 lp/mm at ISO 800 (vertical resolution) is still quite a bit below the expected 53.63 lp/mm. Of course, it is entirely possible that my simulated OLPF effect is not quite perfect yet.
Employing some dodgy extrapolation tells us that the D7000's 4.73 micron pitch is about 25% smaller than the D610's 5.9 micron pitch, so if we add 25% to the 44 lp/mm, we should only be able to get 55 lp/mm out of a D7000. My D7000 measurements came in above 60 lp/mm, and that agrees with other sources (DxO also measured this with the Sigma 17-55 mm f/2.8 lens on this body). That means the D610 + 85 mm images coming from DPReview are not quite as sharp as they potentially could be, but this would also depend on where the resolution is measured (relative to the absolute centre of the lens where my results hail from). At any rate, the lower-than-expected resolution issue is mostly irrelevant to the current discussion, and certainly we are seeing figures that are at least in the right ballpark.
Jack
PS Yikes, it looks like I have been unconventional in my vertical/horizontal naming for months!
