This is part of a series about developing a quick qualitative lens tilt test. The series starts here.
I thought more about the target for the lens tilt test, and realized that I had the wrong one. The target that I was using was optimized to give large amounts of image contrast over a range of defocusing. That’s precisely the opposite of what I want. I want a target that gives constant contrast across its extent, and one whose contrast varies highly with the amount of defocusing.
I thought that a simple checkerboard would be good. I created one, and added a center indicator in a color that would contrast with any of the Sony a7x focus peaking colors so that the rotational accuracy could be assessed.
I removed the Vello adapter that I suspected of creating the tilt that I observed yesterday in the Nikon 70-200 f/2.8E test, and replaced it with a dumb Novoflex adapter. Now the lens had one usable aperture: f/2.8. I set it to 135mm for this test.
Then I looked at the back of the a7RII for two lens rotations 180 degrees apart from each other:
The camera wasn’t perfectly square to the target, as indicated by the dearth of redness on the right side of both images. But the less colorful areas are on the right side of both images, indicating that the lens is not tilted. If it had been tilted, the less red area would have moved as the camera was rotated.
Here’s the same test with the lens set to 200 mm:
Unsurprisingly, the conclusion is the same. The lens is not tilted.
I performed a similar test with a Sony 70-200 f/4 FE lens set to f/4. Here’s what I saw:
The bottom image is the one I used when I lined up the camera. The top is what I saw after rotating the camera through 180 degrees. There is a lot of off-axis falloff in sharpness. That’s due to a combination of field curvature and the fact that this lens does not produce very sharp corners wide open at 200 mm. I’m guess more of the latter than the former, and I think I can prove that. If most of the softness were due to focus curvature, then there would be some position of the focus ring that showed a red doughnut-shaped pattern. There is not.
The top image does not show the same radial symmetry, indicating that the lens is tilted slightly in the left/right direction. I have tested this particular lens previously and found the tilt not objectionable, so this test is pretty sensitive.
I should note that the results for the Sony zoom are not as repeatable as those for the Nikon lens, since there is a bit of play in the Sony removable lens collar.
It would seem to me that a finely demarcated checkerboard is an ideal mechanism to determine rotational alignment of the lens/camera. It should also be equally effective in judging lens axis alignment; defined as being at right angles to the plane of the target where that target is centered in the optical axis of that lens In that case, some form of keystoning (non-rectangular behavior) would be obvious if the target is not ‘square’.
I have NOT done the math, but intuitively any off axis alignment should be immediately visually apparent via non-parallel lines. At least any past testing I’ve done seems to indicate this. Even if geometric distortion is present, the corner endpoints should define parallel lines if the lens is on axis with the target.
For example, the Nikon test shots appear to be square. If we count the displacements in the horizontal and vertical rows of squares in the outer margins it appears to be the same in their given H or V orientation. Those 4 points determine two sets of parallel lines. As an aside, I don’t see how any distortion or alignment could produce a non-rectangular parallelogram. That might be intuitive to me that the target must be aligned ‘square’ but your results seem to indicate differently and I’m unable to rationalize that.
(I have to assume that the non-rectangular screen shots are due to the viewfinder capture mechanism and not the actual test file itself. ??? )
Bruce
That’s right. I just snapped pictures of the LCD with a hand held D5.
I think your comments on the geometry of the target and lens distortion have the ring of truth to them, but I don’t know enough about lens design to say for sure.
Note that having the lens axis perpendicular to the target is not necessary for this technique to find a tilted lens. In fact, the only instance in which the lens will lend up perpendicular to the target in the initial alignment is if it has no tilt.
If we couple that with what you said about the geometry, it may be that there’s a test that doesn’t require inverting the lens. Get the focus peaking radially symmetric, then look at the checkerboard.
Hmmm…
Jim
Hi Jim,
I have implemented a camera calibration process inside of MTF Mapper, using circular fiducials to estimate the parameters of the camera. This camera model would implicitly capture the geometric aspects alluded to by Bruce above, since my camera is constrained to be Euclidean (meaning that perpendicularity is preserved) and also projective (meaning that straight lines remain straight). Radial lens distortion is modelled explicitly, so we should end up with a camera model that maps a rectangle (say through four points near the corners of the chart) to a rectangle (in the reconstructed 3D world coordinate frame); the projection of the rectangle onto the image plane then forms a trapezoid from which we can infer the relative angle between the chart and the optical axis.
I posted some early examples here: https://www.dpreview.com/forums/post/58777891
The one thing that I still have to fully grasp is the influence of a tilted lens on the estimate of the relative orientation of the chart (with respect to the camera).
-F