Luminous Landscape Forum
Equipment & Techniques => Medium Format / Film / Digital Backs – and Large Sensor Photography => Topic started by: madlantern on August 10, 2017, 04:13:41 am
-
I have a copy of Hasselblad's XCD 30mm lens for the X1D, and it seems disappointingly soft in the corners. Specifically, the lower right corner is the worse offender. Is this normal, or do I have a problematic lens?
(http://imageshack.com/a/img924/993/f9PI1N.png)
f/5.6 (left side) vs f/3.5 (right side); both center crops. sharpness is about the same
(http://imageshack.com/a/img923/8812/c287J9.png)
f/5.6 (left side) vs f/3.5 (right side); lower right corner. the f/3.5 (right side) is less sharp, but the f/5.6 is still not very sharp
(http://imageshack.com/a/img922/3195/C8MVV1.png)
lower left corner (left side) vs lower right corner (right side), same image at f3.5
(http://imageshack.com/a/img924/1381/xE4nG3.png)
same scene. tilted the camera up and down to get the same building (in the red rectangle) to the lower and upper edges to compare sharpness.
(http://imageshack.com/a/img923/1444/EOmwsU.png)
left side is lower corner, right side is from upper corner. clearly, the left side (lower corner) is less sharp
-
I tried to follow Jim Kasson's guide to testing for decentering with a flashlight.
I don't know what I make of my results. Does it matter if the flashlight is centered perfectly in the frame? I seem to have noticed that even if the flashlight is a small distance from the absolute middle, the bokeh looks different. I inclosed some of those shots to illustrate what I mean.
The shots are at ISO100, f/3.5 and 1/2000 of a sec with a 6 stop filter. Without the filter, it'd have been too bright to see inside the flashlight.
(http://imageshack.com/a/img924/5166/QHIhwO.jpg)
(http://imageshack.com/a/img924/676/VHFYdo.jpg)
(http://imageshack.com/a/img923/3148/bm2fJQ.jpg)
(http://imageshack.com/a/img922/3689/VvywUS.jpg)
(http://imageshack.com/a/img923/6861/Bk6yD0.jpg)
Here's the full resolution jpgs: https://www.dropbox.com/sh/y9fc47xtb5ooudd/AAD0U8LhJxgGeAKeOTz2FMo6a?dl=0 (https://www.dropbox.com/sh/y9fc47xtb5ooudd/AAD0U8LhJxgGeAKeOTz2FMo6a?dl=0)
-
I tried to follow Jim Kasson's guide to testing for decentering with a flashlight.
I don't know what I make of my results. Does it matter if the flashlight is centered perfectly in the frame? I seem to have noticed that even if the flashlight is a small distance from the absolute middle, the bokeh looks different. I inclosed some of those shots to illustrate what I mean.
The shots are at ISO100, f/3.5 and 1/2000 of a sec with a 6 stop filter. Without the filter, it'd have been too bright to see inside the flashlight.
Here's the full resolution jpgs: https://www.dropbox.com/sh/y9fc47xtb5ooudd/AAD0U8LhJxgGeAKeOTz2FMo6a?dl=0 (https://www.dropbox.com/sh/y9fc47xtb5ooudd/AAD0U8LhJxgGeAKeOTz2FMo6a?dl=0)
I've never seen OOF PSFs that look this bad. They look so bad that I think there might be something wrong with your setup. Are you using a one-LED light? Is there some kind of focusing mechanism in the light? How far away are you from the light? I suspect you're using a large flashlight with some beam focusing mechanism because of the black hole in some of the images, which should never happen if this experiment is done right. Also, I've never needed an ND filter, which indicates to me that your light is bigger than what I've been using.
In answer to your question, exact centering should not be important. I turn on the grid lines in the GFX or a7x cameras and use that to center the light, but if the X1D doesn't have them, it shouldn't make much difference with any camera I've tried. It obviously does with yours, though.
I use a one LED flashlight that has a lens diameter of about 1/4 inch. I get about 25 feet away from it. That seems to be far enough, but the farther the better. If you're using a big flashlight like a Maglite, you'll have to get a long way away, like 100 meters.
Jim
-
Sorry to see these results of your lens. IMO clearly a bad copy.
The 30mm has one of the most impressive MTF, I have seen. The 40lp/mm at f 3.5 drops only slightly from 80% to 70% in the corners (infinity). And no improvement at f/8.
My copy of the lens shows this behavior in the centre and all four corners.
-
I have a copy of Hasselblad's XCD 30mm lens for the X1D, and it seems disappointingly soft in the corners. Specifically, the lower right corner is the worse offender. Is this normal, or do I have a problematic lens?
The first set of tests that you showed are sensitive to alignment errors. If you run the numbers with a CoC of 3 um or less, you'll see that they are amazingly sensitive. The outdoor shots, because of the subject distance, are less problematical.
I would suggest using a Siemens Star as a target, however. You'll need to be at least 20 meters away with the 30/3.5.
http://blog.kasson.com/the-last-word/tilt-and-astigmatism-in-the-fuji-32-644/
If you do this test, please note this warning that I gave you in another context: performing this test with a target like the Siemens Star is extremely sensitive; don't expect a good lens to have perfect results.
Good luck,
Jim
-
@Jim Kasson, thanks a lot for your help! Your posts here, along with your blog have been incredibly insightful.
I found the post on your test of the Fujifilm 32-64 with the Siemens Star on the garage door, and tried to replicate it somewhat.I'm afraid I don't have a Siemens Star on hand. I tried something else.
I went outside and shot a clock from about 35 meters (115 ft) away. I aligned the clock in the center, and took a shot at f/3.5. The tripod stayed in the same place, but I then panned and tilted the camera so that the clock was in each of the 4 corners. I took 4 shots. With all the shots, I was on a 5 sec timer, ISO 100, f/3.5, 1/8sec
35 meters should be about the hyperfocal distance per Jim's chart. But I imagine I can't mess up the focusing too badly in any case. The result shows that the left side is much sharper than the right. Does this indicate decentering, or could this be another issue as well?
(http://imageshack.com/a/img924/3876/Qx7CHK.jpg)
(http://imageshack.com/a/img924/5734/l87bkg.jpg)
-
@Jim Kasson, thanks a lot for your help! Your posts here, along with your blog have been incredibly insightful.
I found the post on your test of the Fujifilm 32-64 with the Siemens Star on the garage door, and tried to replicate it somewhat.I'm afraid I don't have a Siemens Star on hand. I tried something else.
I went outside and shot a clock from about 35 meters (115 ft) away. I aligned the clock in the center, and took a shot at f/3.5. The tripod stayed in the same place, but I then panned and tilted the camera so that the clock was in each of the 4 corners. I took 4 shots. With all the shots, I was on a 5 sec timer, ISO 100, f/3.5, 1/8sec
35 meters should be about the hyperfocal distance per Jim's chart. But I imagine I can't mess up the focusing too badly in any case. The result shows that the left side is much sharper than the right. Does this indicate decentering, or could this be another issue as well?
Without good flashlight test results, it's hard to say. It could very well be field tilt, not decentering. Or it could be something else.
I am concerned about the test however. I'm not sure the center is in focus. One advantage in using the Siemens star is that you can check focus accuracy. A lens like the Hassy 30 should be able to create aliasing on-axis on the X1D. If the image is in focus on axis, you'll see that the lines that should converge at the center start to diverge. If the center is not in focus, mild tilt can look worse than it is if it helps sharpness on one side while hurting it on the other.
You can find out if it's field tilt by seeing if you can make the two sides equally sharp by focusing on each corner. If you can, it's field tilt.
Jim
-
Just FYI there are plenty of users on my forum and others using this lens and they all tend to comment on how sharp it is, right to the edges. I think you must have a bad copy or something is out of alignment with your body.
-
(http://imageshack.com/a/img924/5734/l87bkg.jpg)
I downloaded this image and took a close look at it. I no longer think that the center shot isn't in focus, but it's hard to judge because the exposure was so generous for that shot. If the brightness were the same as for the others, it would be easier. I did darken that part of the image to make a comparison, and that helped, but the highlights are too bright to really see what's going on to the degree that I'd like to. Could you post that raw file?
Looking closely at the images, I see that there is pixel-level detail in all four corners, but that the contrast for the right side corners is less. It's difficult to make accurate judgments about pixel level detail in JPEG images since the DCT compression suppresses that. I suggest losslessly compressed PNG files for this kind of thing, or do what I sometimes do on my blog and enlarge the files to 200 or 300% before saving them.
If you can't do the Siemens Star test and/or redo the flashlight test (based on how good the bad corners look, I don't think you're going to see anything with the flashlight test, but you never know), then you should try focusing on the corners and see what happens. If that clears it up, then you have a tilt issue with either the body, the lens, or both, and you should probably send both back to Hasselblad for service. If that doesn't clear it up, and the right side continues to be less sharp than the left and you feel this level of error would be injurious to your photography, then I'd ask that the lens be repaired or replaced.
Jim
-
I would love to do the Siemens Star test, but I can't seem to get one of those test patterns anywhere. There are no results on Amazon for a "Siemens Star". Do I just print one out myself?
As for the flashlight test, I want to do that test too, but I seem to have difficulty following your instructions correctly. A few questions:
1. How far away do I stand? If I stand 20 meters away as suggested, the light is just a small dot that I can hardly make out.
2. How fast a shutter speed should I use? I find that at f/3.5 and 1/2000s, the flashlight is still way too blown to see any detail. So I have to put a 6-stop filter. I'm not sure if this messes with the result, since I couldnt find anyone having to use ND filters previously.
3. Where do I focus exactly? Does it matter? I've read instructions that tells me to focus as closely as possible, and also others that tell me to focus at the flashlight and then just twist the focus ring a little bit so the flashlight isn't perfectly in focus.
4. Does this flashlight work? I've remove its reflective front cap
(http://imageshack.com/a/img922/9772/m9AC46.jpg)
-
I would love to do the Siemens Star test, but I can't seem to get one of those test patterns anywhere. There are no results on Amazon for a "Siemens Star". Do I just print one out myself?
Hi,
I've made a special version of it that's more suited for use with today's digital cameras. It was first introduced here (http://www.openphotographyforums.com/forums/showthread.php?t=13217).
I'm currently updating from my old/obsolete web pages to more up-to-date versions also with better support for Smartphones/Tablets and improved/secure HTML5 support. I'm not finished yet, but the information about various test targets is going to be hosted on a new site that can be found here (https://www.bvdwolf.nl/foto/test_targets/test_targets.html). The second section will take you to the Camera Resolution targets.
Lot's of additional text that better explains how things work, what to look for in the results, and illustrations, have yet to be added, but it's mostly functional as it is. I will be updating those pages as time permits, consider it a Works in Progress. It is part of a larger collection of webpages on technical photographic subjects, but there is, even more, work to do there before I can share a URL with functional page links.
Cheers,
Bart
-
Not sure if I'm understanding everything here, but are you guys advocating to focus with the target centred and then move the camera without refocussing, so as to place the target in the corners?
If so, how do you propose to get anything outwith the first centre shot sharp?
If I am correct here, then you forgot all about the camera trying to make focus in a flat plane, and once you point away at an angle, your subject will always be closer to the camera than you'd hoped.
Please tell me I just misunderstood!
Rob
-
Not sure if I'm understanding everything here, but are you guys advocating to focus with the target centred and then move the camera without refocussing, so as to place the target in the corners?
If so, how do you propose to get anything outwith the first centre shot sharp?
If I am correct here, then you forgot all about the camera trying to make focus in a flat plane, and once you point away at an angle, your focus will always be closer to the camera than you'd hoped.
Please tell me I just misunderstood!
Hi Rob,
Don't know about others, but I always advocate shooting the center with optimal focus, and the individual corners separately each with their own optimal focus. Otherwise one either gets field curvature prohibiting a proper measurement, or plainly different focus distances as you describe.
When the individual corners are each optimally focused, and the difference (which is quantifiable with a Siemens Star like test target) between the 4 corners is significant, then there's something wrong (usually with lens alignment/decentering). A Siemens star can also suggest other causes for reduced resolution, e.g. when the corner resolution is asymmetric (the blurred center is not disk shaped, but e.g. elliptical).
A large difference between center and corner resolution may be a characteristic of the lens design, but one would normally opt for as small a difference as possible (e.g. when comparing multiple copies of a lens, to pick the best one), especially for a lens used for reproduction of a flat plane.
Cheers,
Bart
-
Hi Rob,
Don't know about others, but I always advocate shooting the center with optimal focus, and the individual corners separately each with their own optimal focus. Otherwise one either gets field curvature prohibiting a proper measurement, or plainly different focus distances as you describe.
When the individual corners are each optimally focused, and the difference (which is quantifiable with a Siemens Star like test target) between the 4 corners is significant, then there's something wrong (usually with lens alignment/decentering). A Siemens star can also suggest other causes for reduced resolution, e.g. when the corner resolution is asymmetric (the blurred center is not disk shaped, but e.g. elliptical).
A large difference between center and corner resolution may be a characteristic of the lens design, but one would normally opt for as small a difference as possible (e.g. when comparing multiple copies of a lens, to pick the best one), especially for a lens used for reproduction of a flat plane.
Cheers,
Bart
Yes, that's always ben the huge problem with focussing on your main subject and then reframing for aesthetic reasons. Especially when, as I seem to be doing these days, shooting almost wide open. Maybe we need lenses with really bad curvature of field!
;-)
Rob
-
As for the flashlight test, I want to do that test too, but I seem to have difficulty following your instructions correctly. A few questions:
1. How far away do I stand? If I stand 20 meters away as suggested, the light is just a small dot that I can hardly make out.
2. How fast a shutter speed should I use? I find that at f/3.5 and 1/2000s, the flashlight is still way too blown to see any detail. So I have to put a 6-stop filter. I'm not sure if this messes with the result, since I couldnt find anyone having to use ND filters previously.
3. Where do I focus exactly? Does it matter? I've read instructions that tells me to focus as closely as possible, and also others that tell me to focus at the flashlight and then just twist the focus ring a little bit so the flashlight isn't perfectly in focus.
4. Does this flashlight work? I've remove its reflective front cap
Here are the instructions from my blog:
Put the flashlight at the far end of the room facing the camera.
Turn the flashlight on.
Go back to the camera
Turn it on, too.
Douse the lights
Set the camera's ISO knob to base ISO.
Open the lens all the way.
Focus it to as close as it will focus. That's right, set the camera to the minimum focusing distance.
Center the out of focus image of the flashlight.
If you've got a mirrorless camera, adjust your shutter speed until the zebras go away and/or the right side of the histogram comes off the end about a stop.
If you don't have a mirrorless camera, take pictures, look at the histogram, and adjust exposure until the right side of the histogram comes off the end about a stop.
For other cameras, take pictures and adjust the shutter speed until that happens.
Is the blurry disk all within the frame? If so, you're done. If not, adjust the focus until there's a fair amount of black around it.
I hope those answer your questions about distance, shutter speed, and focus. You will want to do this test in a dark room, or outside at night.
That flashlight looks OK.
-
I would love to do the Siemens Star test, but I can't seem to get one of those test patterns anywhere. There are no results on Amazon for a "Siemens Star". Do I just print one out myself?
Sure. It will have to be at least C-size (17x22 inches) bigger is better. Here's one to print:
http://cmp.felk.cvut.cz/~wagnelib/img/star_full.png
You can also buy one from B&H:
https://www.bhphotovideo.com/c/product/717671-REG/Zeiss_1849_755_Siemens_Star_Test_Chart.html
Jim
-
Don't know about others, but I always advocate shooting the center with optimal focus, and the individual corners separately each with their own optimal focus. Otherwise one either gets field curvature prohibiting a proper measurement, or plainly different focus distances as you describe.
When you do the test from a sufficient distance that the CoC at the corners of the image is small compared to the pixel aperture, you don't need to refocus on each corner. Not refocusing makes this a test for field tilt and field curvature, as well as the other stuff. If you do refocus, you're eliminating both those things, and introducing the chance of differential focus errors which can yield false positives.
If you look here, you can see how far you need to be away for a 33x44mm sensor as a function of focal length and aperture, for a CoC of 3 um:
http://blog.kasson.com/the-last-word/tilt-and-astigmatism-in-the-fuji-32-644/
Farther is OK, but will require a larger target. Closer is not.
Jim
-
You might consider this simple test for lens centering as described by Roger Cicala at Lensrentals:
https://www.lensrentals.com/blog/2012/05/testing-for-a-decentered-lens-an-old-technique-gets-a-makeover/
The main caveats are noted in the article but it's a quick and easy sanity check for newly acquired lenses or in cases when you think your lens might be de-centered. As Jim just posted, you can either print out or purchase a star chart from B&H.
Good luck!
John
-
When you do the test from a sufficient distance that the CoC at the corners of the image is small compared to the pixel aperture, you don't need to refocus on each corner. Not refocusing makes this a test for field tilt and field curvature, as well as the other stuff. If you do refocus, you're eliminating both those things, and introducing the chance of differential focus errors which can yield false positives.
Hi Jim,
I agree, but it does depend on what one is testing. Testing for field curvature, or chromatic aberration correction, spherical aberrations, or lens mount issues, can be done with multiple targets that cover the full field of view for the camera in a flat plane. That still does not exclude the possibility of a non-parallel orientation of the sensor/image-plane to the subject-plane (and even the sensor can be mounted slightly non-perpendicular to the optical axis, and it's usually very slightly rotated as well, within mounting tolerances).
So to test for lens performance, I prefer to test the best possible resolution in all corners separately, because that takes away all other variables. In a well-centered lens, the corners should resolve virtually identical on an approximately parallel image-plane / subject plane setup. If not, then there is already a lens issue. That's before we even consider the other possible issues. But if we do not first isolate the lens decentering we could be looking at several issues combined into one result.
Cheers,
Bart
P.S. Also, the use of high contrast bi-tonal features, leads to exacerbated aliasing artifacts on discrete sampling devices such as our sensors. That is then potentially worsened by Bayer CFA demosaicing that causes false color artifacts because the input signal was not properly high-pass filtered.
P.P.S. The ISO has added the sinusoidal grating version of the Siemens star to their resolution test standards (ISO 12233). Whether that had anything to do with my discussions with Norman Koren at the time, I do not know, but he did add (sinusoidally-modulated) Siemens star charts (http://www.imatest.com/docs/starchart/) to his Imatest software after I suggested it to him, shortly thereafter followed by the ISO embracing it as an alternative to the earlier slanted edge test.
-
Hi Jim,
I agree, but it does depend on what one is testing. Testing for field curvature, or chromatic aberration correction, spherical aberrations, or lens mount issues, can be done with multiple targets that cover the full field of view for the camera in a flat plane. That still does not exclude the possibility of a non-parallel orientation of the sensor/image-plane to the subject-plane (and even the sensor can be mounted slightly non-perpendicular to the optical axis, and it's usually very slightly rotated as well, within mounting tolerances).
So to test for lens performance, I prefer to test the best possible resolution in all corners separately, because that takes away all other variables. In a well-centered lens, the corners should resolve virtually identical on an approximately parallel image-plane / subject plane setup. If not, then there is already a lens issue. That's before we even consider the other possible issues. But if we do not first isolate the lens decentering we could be looking at several issues combined into one result.
Cheers,
Bart
P.S. Also, the use of high contrast bi-tonal features, leads to exacerbated aliasing artifacts on discrete sampling devices such as our sensors. That is then potentially worsened by Bayer CFA demosaicing that causes false color artifacts because the input signal was not properly high-pass filtered.
P.P.S. The ISO has added the sinusoidal grating version of the Siemens star to their resolution test standards (ISO 12233). Whether that had anything to do with my discussions with Norman Koren at the time, I do not know, but he did add (sinusoidally-modulated) Siemens star charts (http://www.imatest.com/docs/starchart/) to his Imatest software after I suggested it to him, shortly thereafter followed by the ISO embracing it as an alternative to the earlier slanted edge test.
Bart, thanks for your insights. The Siemens Star test that I am developing -- and it is still a work in progress -- is an attempt to come up with a lens-screening test that can be performed by someone not skilled in the art of lens testing. The project stems from my belief that many many perfectly good lenses are returned as defective due to poor testing on the part of their would-be owners. Many folks were reporting de-centered lenses, so I took the OOF PSF test advocated by Prof Hank at DPR and tried to make it foolproof. Many people have been successful with it, but not everyone, including the OP in this thread. Every time someone has trouble with the test, I revise the instructions to plug that hole.
But it turns out that most all the lenses that people call decentered aren't. Of the ones that aren't perfectly fine, they seem to mostly suffer from field tilt, astigmatism, and/or local softness. The Siemens Star test is designed to find those things, even in the presence of mild symmetric focus curvature (which, I believe, is most often the result of design decisions rather than poor assembly.
I started out by rejecting all tests that require precise alignment because it's very difficult to do right (I have trouble myself sometimes), it can yield convincing-looking false positives, and I have found that almost no one is willing to do the checks that will detect it.
Your point about the effects of the step function edges in the Siemens Star is wel-taken. I will do some work with an anti-aliased star. However, at this point, I find the pronounced aliasing and false color useful. It is a test for focus error. It is a very fine indicator of astigmatism and coma. Now that I bring up coma, I'm not too happy that the Siemens Star is so good at detecting that, since I believe that it, too is mostly designed in to the lens. But I could be wrong.
With the help of people who I encounter on the Internet, I will continue to try to refine this test. I have already produced charts that tell you how far away you have to be as a function of lens focal length, sensor size, aperture, and allowed CoC. Turns out it is pretty darned far for many lenses and sensors. That in itself may be a problem in the acceptance of the test. We shall see.
I welcome your further comments; I have learned much from you.
Jim
-
I would love to do the Siemens Star test, but I can't seem to get one of those test patterns anywhere. There are no results on Amazon for a "Siemens Star". Do I just print one out myself?
As for the flashlight test, I want to do that test too, but I seem to have difficulty following your instructions correctly. A few questions:
3. Where do I focus exactly? Does it matter? I've read instructions that tells me to focus as closely as possible, and also others that tell me to focus at the flashlight and then just twist the focus ring a little bit so the flashlight isn't perfectly in focus.
May I ask HOW you are focusing? Based upon my experience with the X1D, I strongly recommend that you do your testing by using manual focus in magnified Live View and not moving the camera after you focus. I generally line up the thing I want to focus on in the center of the frame, manually focusing on that in Live View and making an exposure. Then, I reorient the camera so that the thing I focused on is in each of the 4 corners, refocus and shoot exposures. AFAIK, this test should be pretty foolproof in terms of whether the lens is decentered. I have checked my own copy of the XCD 30mm lens. Not only is it not decentered, it is extremely sharp right into the corners.
One other point to note with the X1D lenses when you are evaluating how good the lenses are. Hasselblad X1D files are not "cooked". The default sharpening settings in LR are seriously inadequate. My default settings are Amount=50-60, Radius=.7-.8, and Detail=70.
-
May I ask HOW you are focusing? Based upon my experience with the X1D, I strongly recommend that you do your testing by using manual focus in magnified Live View and not moving the camera after you focus. I generally line up the thing I want to focus on in the center of the frame, manually focusing on that in Live View and making an exposure. Then, I reorient the camera so that the thing I focused on is in each of the 4 corners, refocus and shoot exposures. AFAIK, this test should be pretty foolproof in terms of whether the lens is decentered. I have checked my own copy of the XCD 30mm lens. Not only is it not decentered, it is extremely sharp right into the corners.
One other point to note with the X1D lenses when you are evaluating how good the lenses are. Hasselblad X1D files are not "cooked". The default sharpening settings in LR are seriously inadequate. My default settings are Amount=50-60, Radius=.7-.8, and Detail=70.
You are recommending a different test than I am. I am recommending -- and providing graphs to show -- sufficient target distance so that refocusing at the corners is not necessary if you want the defocus error to be below a certain CoC diameter.
Not refocusing on the corners allows measurement of field curvature and field tilt. Field tilt is one of the reasons most specified for rejecting a lens.
Refocusing on the corners introduces a new error source: differential focus errors. You may be able to do your version of this test and calibrate them out, but I believe that's too complicated for most people. The OOF PSF test with the LED flashlight is much harder to screw up, yet you'd be surprised how many people have trouble with it.
Jim
-
You are recommending a different test than I am. I am recommending -- and providing graphs to show -- sufficient target distance so that refocusing at the corners is not necessary if you want the defocus error to be below a certain CoC diameter.
Not refocusing on the corners allows measurement of field curvature and field tilt. Field tilt is one of the reasons most specified for rejecting a lens.
Hi Jim,
IMHO, the two types of tests do not have the exact same goals. However, they may both achieve one (of several) common goal(s)
If a lens is decentered (which I think is a relatively common flaw, in varying degrees of severeness), then it seems to me that it's impossible to achieve equal focus/resolution in all 4 corners. The shift of one or more lens elements/groups is bound to negatively affect (at least) one corner more than the others. Wouldn't a perfectly centered lens also exhibit equal and symmetrical loss of resolution in all corners?
So, regardless of the method used, either optimal refocusing for each individual corner after repositioning or repositioning without refocusing and depending on DOF, should both lead to the same conclusion. However, refocusing will allow accurate fine-tuning for commonly used subject distances, while depending on DOF may require long subject distances to achieve adequate DOF to be the resolution limiting factor instead of the lens itself. I prefer the lens itself being the limiting factor if the lens is to be tested.
I believe that a Siemens star chart allows the quantification of optimal limiting resolution best in the case of refocusing. While (approximately) rotating the optical axis perpendicular to the surface of the test target, the (more or less) slightly non-plane-parallel orientation of subject plane and sensor plane will introduce some asymmetry in orientation of the limiting resolution, it is still the highest limiting resolution at the four corners that should match (at different angles). The benefit of using the Siemens star chart is that the optimal corner differences can be quantified (and thus qualified as significant or not).
I've personally experienced (and received feedback from others about) multiple cases of subsequently tested lenses with different/quantified corner resolutions (by using refocused sinusoidally modulated Siemens Star charts), before settling on a specific copy. Some of those tests showed superior resolution of a single corner, but different and lower resolution of all other corners And preferred/common focus distance should not play a role in decentering issues, a decentered lens usually remains decentered at all subject distances.
Other tests may allow to demonstrate/quantify other aberrations, but if a lens is significantly decentered, that's going to complicate all other tests for detection of different aberrations. That's why it's the first thing I'd like to eliminate from the checklist, when purchasing a new lens, or checking a compromised lens.
Cheers,
Bart
-
Hi Jim,
IMHO, the two types of tests do not have the exact same goals. However, they may both achieve one (of several) common goal(s)
If a lens is decentered (which I think is a relatively common flaw, in varying degrees of severeness), then it seems to me that it's impossible to achieve equal focus/resolution in all 4 corners. The shift of one or more lens elements/groups is bound to negatively affect (at least) one corner more than the others. Wouldn't a perfectly centered lens also exhibit equal and symmetrical loss of resolution in all corners?
So, regardless of the method used, either optimal refocusing for each individual corner after repositioning or repositioning without refocusing and depending on DOF, should both lead to the same conclusion. However, refocusing will allow accurate fine-tuning for commonly used subject distances, while depending on DOF may require long subject distances to achieve adequate DOF to be the resolution limiting factor instead of the lens itself. I prefer the lens itself being the limiting factor if the lens is to be tested.
I believe that a Siemens star chart allows the quantification of optimal limiting resolution best in the case of refocusing. While (approximately) rotating the optical axis perpendicular to the surface of the test target, the (more or less) slightly non-plane-parallel orientation of subject plane and sensor plane will introduce some asymmetry in orientation of the limiting resolution, it is still the highest limiting resolution at the four corners that should match (at different angles). The benefit of using the Siemens star chart is that the optimal corner differences can be quantified (and thus qualified as significant or not).
I've personally experienced (and received feedback from others about) multiple cases of subsequently tested lenses with different/quantified corner resolutions (by using refocused sinusoidally modulated Siemens Star charts), before settling on a specific copy. Some of those tests showed superior resolution of a single corner, but different and lower resolution of all other corners And preferred/common focus distance should not play a role in decentering issues, a decentered lens usually remains decentered at all subject distances.
Other tests may allow to demonstrate/quantify other aberrations, but if a lens is significantly decentered, that's going to complicate all other tests for detection of different aberrations. That's why it's the first thing I'd like to eliminate from the checklist, when purchasing a new lens, or checking a compromised lens.
Thanks, Bart. It turns out that most of the lenses that people are calling "decentered" are actually not, if passing the OOF PSF test is any indication.
I think that often the problem is one or more tilted elements, which can cause field tilt that will fail my version of the Siemens Star test, and other effects that will fail both our versions, if they are properly performed.
But the key phrase in the above is "properly performed". I don't think that Joe Public can reliably focus on each corner. In fact, I can't do that consistently myself, and I have to make many exposures, focusing in between, to get the reliability that I get with my one-focus test.
By the way, I am experimenting with sinusoidal (I actually coded a raised sine wave -- is that what you want?). With a camera with an AA filter it doesn't seem to make much difference:
http://blog.kasson.com/the-last-word/leica-1803-4-apo-telyt-r-on-sony-a9-field-tilt-curvature-astigmatism/
I'll be testing with no-AA cameras, in particular the small micro-lens GFX.
Jim
-
Thanks, Bart. It turns out that most of the lenses that people are calling "decentered" are actually not, if passing the OOF PSF test is any indication.
Yes, that could be the case.
I think that often the problem is one or more tilted elements, which can cause field tilt that will fail my version of the Siemens Star test, and other effects that will fail both our versions, if they are properly performed.
The Siemens star test is decent at showing asymmetry where symmetry is expected, i.e. in the center of the image with presumed parallel object- and image-planes. An elliptical center star blur can point to tilt issues (maybe even of the sensor itself). Corners show, almost by definition, asymmetrical blurs if astigmatism or coma effects are present, but the blur is reasonably disk shaped on well-corrected lenses, although of lower resolution than the center. But all corners should have similar resolution.
One of the better qualities of these star targets is that focus distance is not critical for the quantification of the limiting resolution, e.g. expressed as % of Nyquist (or pixels or microns if a specific sensor's pitch is used), because that will always produce the same center blur diameter when the shooting distance is not too close. All that changes with distance is the image magnification, but not the resolved limit diameter. Calibration of magnification factor (from focal length and focus distance) is hard, but no longer needed with a star.
But the key phrase in the above is "properly performed". I don't think that Joe Public can reliably focus on each corner. In fact, I can't do that consistently myself, and I have to make many exposures, focusing in between, to get the reliability that I get with my one-focus test.
Yes, it's harder than many think it is. I've been thinking about adding a moiré producing feature as background to the 'star' instead of the medium grey, similar to what I did for my AF micro adjustment target, but the star itself already does make it easier because (in magnified Live-View), one needs to reduce the diameter of the center blur. That's already quite sensitive, and the harder part is often that the focus ring doesn't allow precise enough manual tweaking (most AF lenses have a very short throw). The moiré feature could be a useful tool for achieving plane-parallel orientation.
By the way, I am experimenting with sinusoidal (I actually coded a raised sine wave -- is that what you want?). With a camera with an AA filter it doesn't seem to make much difference:
http://blog.kasson.com/the-last-word/leica-1803-4-apo-telyt-r-on-sony-a9-field-tilt-curvature-astigmatism/
I'll be testing with no-AA cameras, in particular the small micro-lens GFX.
Yes, that's the kind of sinusoidal modulation I use myself. I use a different gamma than in your target so it can be looked at in gamma pre-compensated images, your's are best viewed at linear gamma, although it's not critical if not numerically evaluated for MTF.
The sharper the lens, the more optimal an aperture is chosen for maximum resolution (often around f/4-f/5.6), the smaller the fill-factor, and the larger the photosite pitch, the more you'll avoid excessive aliasing from compromising the analysis if the target produces a well-behaved input signal. It's also more gentle on the demosaicing because it reduces the false color artifacts, thus making e.g. Chromatic Aberration easier to spot.
Your targets use a 60-cycle or 60-sector pairs star, and a 20-cycle or 20-sector pairs stars.
That creates the Nyquist limit blur diameter at an approx. 38.2 pixel, and 12.7 pixel diameter at 100% zoom (2 x cycles / Pi diameter , used as quick estimate).
Cheers,
Bart
-
I'll be testing with no-AA cameras, in particular the small micro-lens GFX.
Today I ran a test with the GFX, the 110/2, using binary and raised-sinusoid stars of 40, 60, and 80 spokes.
http://blog.kasson.com/the-last-word/fuji-1102-on-gfx-off-axis-performance/
I am not convinced of the benefits of the sinusoidal stars if we're going to be judging the images visually, although they would probably help with quantitative machine-controlled evaluation.
Jim
-
Today I ran a test with the GFX, the 110/2, using binary and raised-sinusoid stars of 40, 60, and 80 spokes.
http://blog.kasson.com/the-last-word/fuji-1102-on-gfx-off-axis-performance/
I am not convinced of the benefits of the sinusoidal stars if we're going to be judging the images visually, although they would probably help with quantitative machine-controlled evaluation.
Hi Jim,
Interesting results. Looks like you have quite a decent copy of that lens as far as symmetry is concerned, not perfect but then nothing is.
It looks like the crops were zoomed in, to something approaching 250-300%, so I cannot quantify the absolute system's limiting resolution. When you calculate the Nyquist frequency limit (approx. 51 pixels for the 80-spokes star), you can see how close to Nyquist the resolution gets for real subject detail, and how good the lens is in absolute terms. The choice of Raw converter also plays a role in that chain. You might want to test the center crop with e.g. RawTherapee with the Amaze demosaicing algorithm to see how much of an influence the demosaicing can make.
What the tests also show is how complex the resolution changes are across the image plane. That then also hints at the difficulty of using a single (Capture) sharpening setting for the entire image. Only spatially variant sharpening can get the technically best out of an image (although that may not be necessary from a creative point of view). And as you said, not all subject matter is as critical as these test targets (thank goodness).
Cheers,
Bart
-
Hi Jim,
Interesting results. Looks like you have quite a decent copy of that lens as far as symmetry is concerned, not perfect but then nothing is.
It looks like the crops were zoomed in, to something approaching 250-300%, so I cannot quantify the absolute system's limiting resolution. When you calculate the Nyquist frequency limit (approx. 51 pixels for the 80-spokes star), you can see how close to Nyquist the resolution gets for real subject detail, and how good the lens is in absolute terms. The choice of Raw converter also plays a role in that chain. You might want to test the center crop with e.g. RawTherapee with the Amaze demosaicing algorithm to see how much of an influence the demosaicing can make.
What the tests also show is how complex the resolution changes are across the image plane. That then also hints at the difficulty of using a single (Capture) sharpening setting for the entire image. Only spatially variant sharpening can get the technically best out of an image (although that may not be necessary from a creative point of view). And as you said, not all subject matter is as critical as these test targets (thank goodness).
Thanks, Bart. For the purposes of lens screening, the absolute resolution is not that important, but I like to see aliasing on-axis as an indication of correct focus. When you get "pillows" (as Jack Hogan calls them), then you know for sure you're focused to the level the sensor can resolve, but not all lenses can produce the pillows that the 110/2 does.
Brandon Dube has sent me some simulated Siemens Stars for two common lens aberrations for a monochromatic sensor. Eventually, I'd like to create simulations for a Bayer CFA sensor, to show people (and me!) what those aberrations look like.
My worry now is that the test is so sensitive that it will show evidence of lens issues that will have no effect whatsoever on normal photography, and may result in good lenses being returned. From that point of view, the raised-sinusoid star is a better target, I think. I need to work on screening criteria.
Jim