Hi Bart,
That's a really great printer target you have there. I've been going along believing, incorrectly it turns out, that if an image doesn't have frequencies close to Nyquist at 300 DPI then nothing was to be gained by running the Canon with a higher DPI image. That clearly is not the case. Even when the image has frequencies that can be fully resolved at 200DPI there is some aliasing when printing at 300DPI and hence there is a benefit to upsampling even if no sharpening is done.
Hi Doug,
Indeed. Even before image detail reaches the Nyquist limit, it takes very little to generate what seems like moiré but technically can't be. It's the quantization and gamma effects that cause that apparent moiré, but even if it is technically not aliasing, it can still look like it and distract in our images. The gamma effect can already be see on display when you look at the finer detail of my target at 100% zoom, and move your head up and down in front of the display. The minute change in gamma will generate patterns where there are none.
Oversampling is the best way to prevent as much of that as possible, and since it's cheap to do with printing (by switching on 600 or 720 PPI), I recommend making it the default practice.
While the printer resolution has not been something I've thought about much I've been very interested in measuring sensor aliasing and developed a similar target of concentric sine waves for investigating this. The target is quite large and LPI very low but shot at a distance it is easy to determine optimal aperture on various cameras. For instance F22 on a 1DsIII and F16 on a 5Ds R both are happy spots where color artifacts from CFA aliasing is minimized. Twice as large an aperture works great on B&W "document mode" imaging using dcraw and removing the CFA effects. At least on lenses with low chromatic aberration.
Correct, at those apertures the MTF for green wavelengths/luminance reaches a response of zero at the Nyquist frequency, just like the
theory of circular aperture diffraction with a very good lens predicts. You'll probably get almost similar results at 1/3rd a stop wider, due to residual lens aberrations that blur the image slightly.
The role of the Rawconverter is also a factor of importance. Some demosaicing algorithms extract more resolution from an image than others. Capture One, RawTherapee (with Amaze), and Iridient RAW developer (=Mac OS only) all have a good reputation of high resolution with a minimum of artifacts.
Another target is made of concentric patterns of 7 overlapping sine waves each of which has the same amplitude. Running FFTs on that yields excellent MFT responses correlating nicely with camera motion from mirror slap and first curtain mechanical induced movement. These cause a kind of low pass filter effect along a motion vector. At least to first order.
I'm also investigating simple/useful tools to assist in quantifying vibration blur. I believe you've posted (here on LuLa) about that method before, but I wondered how the varying magnification factor from focal length and shooting distance was calibrated. As camera resolution increases, so does the visibility of camera shake.
But I've been ignoring the printer! When I was printing some 5Ds R images on smallish (6x9) sheets I was initially downsampling to 300 DPI thinking that it would reduce aliasing and noticed some odd results in fine, repetitive patterns that were not there in the original 5Ds R image. So I took my targets used for sensors, downsampled 10x, and printed them but with freqs still below Nyquist and was surprised to see aliasing (Moire).
Paper choice also makes a difference, which is what sparked my interest in designing such a target. Some papers exhibit more ink diffusion than others. But with well behaving paper coatings, it really pays off to print at the highest resolution (600 or 720 PPI) that the printer allows. It also helps with output sharpening.
Now I see you had explored that road in depth!
Very interesting discussion in your link. Good stuff Bart.
I think it's always useful to share such stuff, so you're welcome
Cheers,
Bart.
