Printer Resolution and Stuff

PPI is sort of a requirement (close to a hardware requirement). For a given printer / imagesetter / media (paper, film, ...) / desired printing speed / desired quality: ppi is a constant.

And there is no preview, which, for an imagesetter, meant extra Cromalin proofs.

Yes, the dialogue box in Photoshop Elements is the same and so you can choose whether or not to resample to the printer's native print resolution.

But in this case you are limited to the resampling type options in PS/PSE.

I wonder if the modern AI driven resampling apps are even better than the PS/PSE algorithms. If they are then the resampling process is not as simple as simply checking/unchecking the "Resample Image" checkbox.

Qimage has good resampling.

I'm sure it does but I also wonder if the modern AI driven resampling apps are even better.

I wouldn’t use them for routine resampling, since they are prone to generating artifacts. If I had time to go over the resampled image with a fine toothed comb, then they are worth a hard look.

I find they work very well for me most of the time with maybe just minor edits to fix anomalies in the output.

I agree. That's not different than what I said above. But I'm having a hard time squaring your desire to get even better quality resampling than Bicubic or Lanczos with your statement that nearest neighbor is good enough for most things. Maybe I'm oversimplifying what you've said in this thread.

Earlier this thread I posted when generally the printer driver interpolation is good enough for me and when I would do the resampling myself manually. For those cases I posted I use AI driven apps to do the resampling.

Yes and in fact the Slanted Edge Method does exactly that: it effectively reconstitutes the average 'continuous' profile of the edge, shot noise and all, keeping in mind that the 'continuous' signal is quantized to begin with per Planck and Einstein:

The quantized signal can be turned into a 'continuous' one in many ways. I like monotonic regression per Kruscal (step) or de Leeuw (spline), I can't remember which of the two was used to generate the thin orange line that can be seen under the blue dots in the figure above. That's the Edge Spread Function, which as we know is a 1D projection of the cumulative mean 2D PSF onto the edge normal. If desired, it can then be transformed into a Modulation Transfer Function, which is a convenient interpretation of the ESF in certain applications.

Such an MTF represents the average loss of contrast at various spatial frequencies for the hardware tested as setup, in the center of the edge and in the direction perpendicular to it. Pick a threshold loss of contrast (e.g. 50%, 9% or whatever works for your application) and you have a single figure 'working resolution'.

So the key job of the Slanted Edge Method is to estimate the continuous ESF, which it performs admirably. So well in fact that the resulting continuous ESF has lost all information about the original sampling interval (pixel pitch), hence aliasing. It contains all other relevant spatial frequency information though, including the impact of the lens, the filter stack, microlenses and pixel aperture.

It leaves it up to the operator to estimate beyond what frequency the MTF curve could become less relevant as a result of aliasing. For instance, CPIQ Acutance, a sharpness measure used by smartphone camera manufacturers, takes into consideration a weighted average of the curve up to monochrome H/V Nyquist (Nyquist changes with the color of the detail and its angle to the sampling grid). Others tend to be more stringent.

It would be interesting to see some work quantifying the perception of aliasing in various common conditions.

Jack

PS More detail in this article: www.strollswithmydog.com/the-slanted-edge-method/

A sin wave with a steadily increasing frequency must, if it's a physical phenomenon such as a sound, have a physical limit. An image made up of pixels is limited by the size of the pixels. An image made of ink is limited by the size of the molecules.

Don

Have not find one that I can wholeheartedly suggest yet.

I have not seen any. I think this is a good case to AI.
"Could you please print this photo with my printer in size of 40x50cm. The best possible image quality.
Crop to aspect ratio."
Or, "Don't crop, I'll do it by myself."

So, that is specific to that printer and driver?

So, the method of dither is changed; not just the size of the dithered virtual pixels?

I've done some more research on this, and the defeating of the variable dot sizes with Epson printers is based on the marking engine resolution chosen, not the enhanced details setting. So I was wrong about that, at least as of about five years ago. And, yes, it is manufacturer-specific.

That's OK. There are plenty of recommendations on the www based on pros and cons of each.

Or people can do their own evaluations based on their needs.