Towards a quick, qualitative lens tilt test

In this thread on DPR, the original poster reported some creative and useful work on quantitatively measuring filed flatness and tilt. During the discussion, he made reference in passing to using the focus peaking feature of some mirrorless cameras to rapidly check for lens tilt. 

I was intrigued.

In the past, I had some success in quickly checking for decentering using a technique developed by a University of Kentucky professor and publicized on DPR. However, that method is insensitive to tilt. That’s a good thing if you wish to separate the measurement of tilt from that of decentering, but not such a great characteristic if you just want to quickly check a lens to see if it’s a good copy.

I had a new Nikon AF-S Nikkor 70-200mm f/2.8E FL ED VR that I wanted to test. First, I tested for decentering.

Well, that looks outstanding. I do see a big ugly dust spot in the 200 mm image, though — it looks a bit like a sunspot.

For my tilt test, I used a target that Horshack, another DPR contributor, had given me:

My objective was to come up with a way to detect field tilt that required no measurements or special alignment gear.

I printed out a 17×22 inch version of the target, and mounted it on door:

I set up a Sony a7RII adapted to the Nikon zoom with a Vello smart adapter, and mounted the combination of an Arca Swiss C1 cube:

I set the lens to a focal length of about 135mm, and filled the screen with the target. I set the focus peaking to low, which is very strong for this application. If the image is at all close to being in focus, the screen lights up uniformly, which is not useful:

However, by twisting the focus ring on the lens and defocusing, the setup can be made exquisitely sensitive to focus errors. Note that we’re not actually looking at an image where anything is perfectly focused, but rather one where we can measure relative blurriness of a slightly OOF image. 

How do we ling up the camera? Turns out that it’s easy. Just move the camera back and forth and up and down until everything is pretty even, then slide the tripod over to a place where the target fills the frame with the camera pointing in that direction. If there’s tilt to the lens, that won’t necessarily put the mechanical lens axis in the middle of the target, but it will turn out that that won’t matter if we’re just looking for the existence of tilt. 

Let’s say we line the lens up so the target looks like this:

Then we rotate the camera a80 degrees in the tripod collar, and look again:

This image is now sharper on the left side of the inverted screen. That means that there’s some tilt in the system. Since the camera’s foucs distance was short of the actual target distance in this case, we can see that the right side of the (non-inverted) image focuses a bit closer than the left. 

Note that the target is not quite square in either image. The rotational difference is not quite 180 degrees. That’s no problem because we’re looking at differences across the frame.

It works in portrait mode, too.

Now we see that the right side — as seen by the camera — focuses closer than the left. 

Some caveats:

I note that the Sony a7x focus peaking system only looks at horizontal changes. However, tilt in any direction should produce softness as measured side-to-side, so this should work for tilts in any direction. I need to test that.

There is a problem when you test adapted lenses this way. If there’s a tilt found, it’s not obvious if it’s the lens or the adapter.

There us also the possibility that this test is so sensitive that, should it become popular, that perfectly good lenses might be sent back.

It should be possible to extend this approach to testing lenses without focusing collars by using a mounting arrangement that lets the lens rotate about its optical axis. In the device that I have for performing this operation, the alignment of the lens axis with the rotational axis is manual. Fortunately, it’s easy to see how good the alignment is by rotating the lens and seeing if the central point of the target moves. If you turn on the finder grid on the a7x, that point is precisely marked. 

Testing wide angle lenses with this method will require large targets.

Testing field flatness should be pretty easy in spite of the fact that focus peaking only works in one direction, since field flatness errors should, to a first order, affect most directions equally.

I’ll do some more work and report here.