Simulated architectural shots vs fill factor

Yesterday I posted a series of images of foliage as captured by a simulated camera with fill factors varying from 1% to 100%. The aliasing artifacts of the low-fill-factor images were remarkably hard to spot. Today I’m going to do the same thing with a more demanding subject. Here it is, downsampled for the web:

I called this “architectural” in the title of this post, but it may not of the class of images that you first think of when you hear that term. It was made many years ago in Siena, Italy, with a Leica M8 and a 50mm f/1.4 Summilux ASPH. I set the simulator for a 5.3 um pitch camera with a diffraction-limited lens, and used AHD demosaicing. Here’s what I got, at a very small part of the center of the simulated sensor, at f/2 for both 100% and 60% fill factor, enlarged to 400%:

To my eyes, the differences are remarkably small. There is obvious stair-stepping in both images, indicating that our diffraction-limited f/2 lens (try finding one of those at B&H) is way ahead of our sensor.

Now let’s look at fill factors of 10% and 1%:

Both of these show obvious aliasing. Look at the two sets of light-green shutters in the center of the picture. They show dark sections that aren’t really there.  But they also show some real information that didn’t show at the large fill factors: look at the vertical parts of the railing in the lower left. Note that some of the vertical bars are missing entirely. The texture of the wall on the right is enhanced at low fill factors, where it was suppressed at high ones. Whether this is good or bad will depend on your tastes.

I have been asked if most or all of the difference between 100% of the X1D and the GFX’s lower fill factor could be eliminated with deconvolution sharpening. My initial impression based on experiments with the Nikon D800E and D800 (which did not have different fill factors, but has different approaches to anti-aliasing) is that most of it could be. But it’s also clear to me that no amount of deconvolution sharpening is going to make the vertical parts of that balcony railing in the 100% fill factor image appear as they do in the 1% image. On the flip side, no amount of deconvolution sharpening to the a00% image is going to reproduce the artifacts we see on the green shutters in the 1% image.

Now let’s look at what happens when we stop our simulated lens down to f/4. This is representative of what lenses that you might actually be able to purchase san do on-axis.

Again, there is not a lot of difference between the 100% and the 60% images.

The green shutter aliasing is still present in the 1% and 10% images but is diminished in intensity. The vertical parts of the railing are visible in the two low-fill images and not in the two high-fill ones. They look more natural now that the lens in introducing a bit more blur.

At f/8:

Now the differences are virtually gone.