Siemens Star “pillows” — part II

This is a continuation of the development of a simple, relatively foolproof, astigmatism, field curvature, and field tilt test for lens screening. The first post is here.

In the previous post of this series, I simulated the “pillows” that appeared in the Siemens Star images made with the Fuji GFX 50S and the Fuji 110/2 lens. I used two demosaicing methods which are fairly unsophisticated compared to the algorithms employed by raw developers like ACR and Lightroom  (Lr). I wanted to see what happened to the simulated images using ACR.

I had some Matlab code that Jack Hogan had written to paste an arbitrary image into a DNG file from a real camera. His idea, which became my idea, was to paste mosaiced images from my simulator into a file from a real GFX 50S, and develop the resultant image in Lr.

I set the simulator to make an image from a camera with a 5.3 um pitch, 60% fill factor, and a diffraction-limited f/2.8 lens. I guessed at the CFA compromise matrix for the GFX. Here’s a magnified look at what I got when I developed the result in Lr:

The real image is on the left part of the picture. The simulated one is on the right. I told the sim to write out all possible CFA patterns in case I got my math wrong, which is why there’s that ugly magenta thing on the bottom right. The one with the right orientation {RGGB, for the GFX) is on the top right. The cream-colored background indicates that I got the white-balanced compromise matrix coefficients a bit wrong, but that shouldn’t hurt anything. When Lr developed the above image, it didn’t know which parts were simulated and which were real, so, if the demosaicing algorithm were the long pole in the tent, the 60-spoke simulated star on the right would look like the real one that it partially occludes.

You can see that the real 110 mm f/2 lens used at f/2.8 is not as sharp as a diffraction-limited f/2.8 lens. That shows up in the greater overall contrast of the simulated image, the smaller area in the center that is completely unresolved gray, and the more intense false color. However, the size of the false-color “pillow” is about the same in the two images, indicating that, for full-color resolution, the Fuji 100/2 at f/f.8 is sufficiently sharp that the sensor is the limiting factor.  The shape of the pillows is basically the same in the real and the simulated star renderings. 

Because of the way the Siemens Star is constructed, it shouldn’t matter how big or small it is in the image, as long as there is enough detail in the star to have a flat unresolved center and a fairly high contrast periphery.  Just to check that, I set the simulator for the diffraction-limited f/2.8 lens, 60% fill factor, a 5.3 um pitch, and fed it a number of different sized stars, starting with a large one at the top left of the image below, and ending with a small one at the bottom right. I used AHD demosaicing. I scaled all the stars back to the same size in this image:

Except for size, the differences are small.