Am I getting this right?

[perdu34]

I've been doing extreme macro for about a year now and I have a grand total of 13 images I am happy with and many many more that I'm not happy with. Of those 13 only 2 or 3 are really up to the standard I really want them to be at.

With that in mind I'm gonna outline my entire workflow below and maybe someone can point out where I am going wrong?

Thanks in advance




Camera Set up

I have a Canon 5Diii with a Canon MP-E 65mm set to 5x and I have the twin flash as well as two yonguno flashes. Each flash head is set at 90-degrees and roughly balanced to give as even an illumination as possible.

Around the subject is a 3D printed diffuser and shade. The shade is mean to make the background as black as possible and as a silver foil covered faceplate so that it gives so backlighting.



My typical stack settings are f4, 1/160s, ISO 100. The camera is on a Stack shot and takes a set forward, puts the mirror up, waits 10s and then fires the shutter. It then moves forward 10-15um depending on how detailed I think the subject is (hairy subjects get bigger stacks). Once a stack is done I typically move to an overlapping area of the subject and stack again. A normal 20-30mm beetle will have 30-60 stacks.

As I acquire onto a 2tb portable hard drive and process on a different computer (in another part of the city) I acquire all the images and then process as a batch.

Stacking

The first thing I do is go into Lightroom and process the RAW files. Normally this is just making sure the histogram is fully used and there are no clipped highlights. I sometimes bump the contrast and clarity as well as add a little vibrance if the bug is very bright and colourful. These get batch exported as jpgs with the Adobe1998 colour space.

I have played with Helicon a lot and found that under almost all circumstances I get good stacks with Method B and a radius of 10 (with the MP-E 65mm at 5X). Smoothing will change a bit but not much.





Stitching
Once all the images are stacked and sorted I load them into Photoshop for stitching. I use the load into stack function and make sure the auto-align box is unticked.


I will then do a quick auto-align just to see what fits and what doesn't. Normally its a leg (or 6) that doesn't fit right.



Knowing which limbs don't fit I go around removing as much background and stacking artefacts from those images and re-align. This can take forever. And I am a Mac user so I can't use ICE which I am told does a way better job that Photoshop.


I then use the auto-blend function to make sure that subtle changes in colour are accounted for and fixed.

Photoshop and Editing
I will then go around the image and highlight all the bits I am unhappy with. In this case I broke a leg off while imaging, there were lots of stitching errors and the pin needs removing. There were also a few stacking issues I hadn't noticed before.


After I've died a good number of those issues I will then use the pen tool to cut out the subject from the background. It seems odd that I go to a lot of effort to make sure the background is as black as possible only to then cut it out in Photoshop but it just looks cleaner than using levels or curves and selections to bring a very dark grey background to true black. To me anyways.



Next I go around with the clone stamp and heal brushes removing any dirt and other imperfections in the image. Some times I see stuff like this where a mite is living on my subject.


This cleaning process can take a good 10-20 hours in photoshop. I am getting faster at it but the more I do it the more dirt I see. I know no one will ever look at these images at 500% crop and go "theres a speck of dirt on that leg" but I do.

The image will then get a final crop and colour and contrast fixes in the same way I would with a normal image.







If you guys know where I can improve, what I'm not doing right then please let me know.

The things that are really bothering me are:
Lack of sharpness
Stacks are less sharp overall than the in focus sections of the unstacked images
Poor stacking (lots of artefacts all the time)
Poor stacking around hairs.
Hotspots at the top and bottom of the bug

[mawyatt]

I've had difficulty stitching with PS and Rik pointed me to PTGui Stitching software.

https://www.ptgui.com


This is much better than PS. I haven't purchased it yet, but did the demo on a problematic image set that PS was struggling with, PTGui breezed thru the set without issue.

As soon as I have another stitching session (coming up soon) I'll be purchasing PTGui.

As far as a sharper image goes the MP-E 65 isn't considered that sharp at 5X according the others, I don't have one (Nikon based) so can't say from experience. The Mitutoyo 5X M Plan App Microscope Objective is considered very sharp at 5X, and is quite good (have one).

BTW your image looks good and I like your diffuser & shade (please tell us more about this)!!

Edit to change to PTGui.

Best,

[rjlittlefield]

Mike, I'm wondering if something in your records or your typing has glitched.

According to my email archives, it was PTGui (not Hugin) that I recommended to you, and that you reported success with.

PTGui is also the one that has to be purchased; Hugin is open-source freeware.

--Rik

[JH]

Hi perdu34
Verry nice picture and informative and interesting description if your work process.

Stacks are less sharp overall than the in focus sections of the unstacked images

Usually the stack is as sharp as the source images- so this is a problem that I can not explain. Do you by any chans stack lower resolution (down-scaled) pictures?

For pictures with much higher resolution than the mp-e 65 you have to use a microscope objective with 10x na 0.30 (0.28 ) and aprox. 6x more pictures in every stack. Bevare that you need good corners for stack and stitch and the image circle of microscope objektives is quite small.

If you have problems with hairs you could try for exemple 3x and a higher f value than the 2.8 you probably use for 5x.

Best regards
Jörgen Hellberg

[mawyatt]

Mike, I'm wondering if something in your records or your typing has glitched.

According to my email archives, it was PTGui (not Hugin) that I recommended to you, and that you reported success with.

PTGui is also the one that has to be purchased; Hugin is open-source freeware.

--Rik

Rik,

Yes, thanks for catching this. My bad

I've edited to PTGui, hopefully I didn't steer anyone in the wrong direction!!

Best,

[rjlittlefield]

Lucas, I think your basic problem is that you're trying to stitch images that don't actually fit together properly. The reason they don't fit together properly is that stacks shot with the MP-E 65 lens will inevitably have quite a bit of perspective distortion. The field of view of the lens fans out from a center of perspective (the lens's entrance pupil) that is not very far inside the lens. When you slide the lens sideways to shoot the next stack, the lines of sight for corresponding features are not parallel, so you end up with parallax errors. It is exactly the same problem that arises in stitched panoramas when the camera is rotated around any point other than the magic "no-parallax point", which in fact is the center of the lens's entrance pupil.

To solve this problem, and to maintain full sharpness, you need to switch to "telecentric" optics.

The best approach is to buy lenses that are specifically designed to be telecentric, such as the Nikon Tool Maker Objectives (EDF20031, 3X NA 0.09, 75.5 mm WD; EDF20052, 5x NA 0.13, 64 mm WD). Obtaining those objectives may require some discussion with your local Nikon representative. They are not terribly expensive, around $600 and $700 a couple of years ago, but Nikon may want some extra paperwork because they are not normally sold in quantity 1 to individuals.

Given the telecentric optics, you should then turn off scale correction in your stacking software. If your stacking setup is stable enough to permit it, you should also turn off shift corrections. Using the combination of telecentric optics, stable setup, and no shift or scale corrections in software, the stacking process will end up consisting of little more than plucking pixel values from source images and assembling them into an output image. In that case there would be no subpixel sampling, so no opportunity for softening -- assuming that the stacking software correctly chooses the best-focused frame at each pixel position.

There is, however, an issue with having the stacking software correctly choose the best focused frame at each pixel position. You mentioned that with Helicon you get good results using Method B (depth map) with a radius of 10. That combination is excellent for smooth surfaces such as your beetle's elytra, but it's pretty much guaranteed to produce significant artifacts at places where features overlap but at different depths. The best handling in those areas will probably come from a pyramid method, which in Helicon is Method C. Then you'll have to retouch those areas from the pyramid output into the depth map output.

Depending on other issues that can be hard to predict, pyramid and depth map often produce different average brightnesses for the same area. Historically this has made retouching in Helicon a bit problematic because its retouching brush does not know how to correct for different brightness so it's easy to end up with light or dark streaks. I don't know if this is still an issue in Helicon's latest beta. In Zerene Stacker, the brush is specifically designed for retouching from PMax (pyramid) into DMap (depth map), and in particular it automatically adjusts for brightness differences so that the retouching is seamless. (See HERE and in the surrounding thread if you want more discussion about that.)

One more comment about the stitching. I'm assuming that your setup simply slides the field of view across the subject and does not rotate the camera. If that's correct, then the image as stitched should also show a regular mosaic of rectangular frames. In contrast, the fifth image of your post (HERE), shows an prominent "hourglass" layout of the tiles, with tiles being distorted into keystone shapes and rotated away from horizontal. The effect of that will be some shape distortion of the subject -- an artificial widening of the posterior in this case.

--Rik

[cube-tube]

Rik, as a rule of thumb, are all objectives labeled as "toolmaker" objectives telecentric?

[rjlittlefield]

Rik, as a rule of thumb, are all objectives labeled as "toolmaker" objectives telecentric?

Sorry, I don't know.

Given the choice between an objective labeled "toolmaker" and one that is not, surely I would grab the "toolmaker" to test first. But the emphasis is on "test", and "first" is in there to indicate that I would not be surprised if the test said nope, not telecentric.

By the way, there's also an issue that lenses can be telecentric over just part of their field. So, an objective that is telecentric on APS-C might not be telecentric in the corners & edges of full-frame.

--Rik

[perdu34]

BTW your image looks good and I like your diffuser & shade (please tell us more about this)!!
Best,

Hi Mike, there's some info on the diffuser (and the rest of my set up) here

There have been some minor adjustments since then but I am happy to share an STL file if you have access to 3D printers.

[perdu34]

Lucas, I think your basic problem is that you're trying to stitch images that don't actually fit together properly. The reason they don't fit together properly is that stacks shot with the MP-E 65 lens will inevitably have quite a bit of perspective distortion. The field of view of the lens fans out from a center of perspective (the lens's entrance pupil) that is not very far inside the lens. When you slide the lens sideways to shoot the next stack, the lines of sight for corresponding features are not parallel, so you end up with parallax errors. It is exactly the same problem that arises in stitched panoramas when the camera is rotated around any point other than the magic "no-parallax point", which in fact is the center of the lens's entrance pupil.

To solve this problem, and to maintain full sharpness, you need to switch to "telecentric" optics.

It's my understanding that the "correct geometric distortion" function in photoshop is meant to fix this kind of problem as they are coming with normal panoramas too. Am I wrong?

The best approach is to buy lenses that are specifically designed to be telecentric, such as the Nikon Tool Maker Objectives (EDF20031, 3X NA 0.09, 75.5 mm WD; EDF20052, 5x NA 0.13, 64 mm WD). Obtaining those objectives may require some discussion with your local Nikon representative. They are not terribly expensive, around $600 and $700 a couple of years ago, but Nikon may want some extra paperwork because they are not normally sold in quantity 1 to individuals.

I will have to look into telecentric optics again. I had ruled out at some point as I was under the impression that a normal 10x NA0.3 was close enough to telecentric as to not make too much of a difference. I am in the process of getting a 10x set up built but it's on the slow burner as I wanted to get 5x to a place I am happy with it.

Do you think that moving to a 10x set up (Olympus 10x FLPlanU NA0.3 and a Pentax 200mm f4 tube lens) would inherently improve the stitching? This combination gives me a full FOV on my Canon.

There is, however, an issue with having the stacking software correctly choose the best focused frame at each pixel position. You mentioned that with Helicon you get good results using Method B (depth map) with a radius of 10. That combination is excellent for smooth surfaces such as your beetle's elytra, but it's pretty much guaranteed to produce significant artifacts at places where features overlap but at different depths. The best handling in those areas will probably come from a pyramid method, which in Helicon is Method C. Then you'll have to retouch those areas from the pyramid output into the depth map output.

I have been playing with helicon's brush features and I've not been too impressed, which is why I use photoshop more. Zerene is something I am starting to play with now, I do like the brush tools but I am still very new to it. I will give the tutorials a go and check out the link you sent.

One more comment about the stitching. I'm assuming that your setup simply slides the field of view across the subject and does not rotate the camera. If that's correct, then the image as stitched should also show a regular mosaic of rectangular frames. In contrast, the fifth image of your post (HERE), shows an prominent "hourglass" layout of the tiles, with tiles being distorted into keystone shapes and rotated away from horizontal. The effect of that will be some shape distortion of the subject -- an artificial widening of the posterior in this case.

--Rik

I do believe I may have rotated the camera between imaging the legs and the body, but I can't remember now (and I don't have access to the original images at this moment as they are on my other computer). Its not something I usually do but it does save time as it increases overlap in the limbs as they are essentially "portrait" in shape.

[cube-tube]

It's my understanding that the "correct geometric distortion" function in photoshop is meant to fix this kind of problem as they are coming with normal panoramas too. Am I wrong?

Geometric distortion and parallax are two different things, and Photoshop will probably do a good job with distortion, but it can only make a crude attempt to fix parallax. It is true that parallax is often a problem with normal panoramas as well, but photographers get around that by panning the camera around the nodal point with a special panorama rig.

I would be surprised if parallax is causing real problems for you, though. Parallax is only a problem when you have foreground and background objects. You might have some problems with some of the legs or the head, but the majority if your subject is a continuous, smooth surface.

I know that Levon Biss, who used a method almost identical to yours to make giant insect portraits, did not use perfectly telecentric optics, and he had good results even with insects with fairly complicated geometry. It does take a lot of work retouching, though, so telecentric optics might just save you some time.

You can also simulate telecentric optics with your stacking software. After all, you are moving the camera/subject for each frame, so everything (foreground, background) should appear at the same magnification if the software doesn't do any scaling or other adjustments. I'm a little fuzzy on that idea, though, and I have never tried it myself, so don't take my word for it.

[anoldsole]

If you intend on doing a lot of stack and stitch, I highly suggest using telecentric optics like the nikon telecentric objectives. It seems you are trying to optimize your workflow, and you will spend less time cleaning up the stitching artifacts if you do.

For the price i'm not aware of a better telecentric option. I think the mag.x has superior image quality, especially on a full frame, but us mere mortals will have to be happy with the nikons. A 3x will not cover a ff without vignetting but a 5x is just about acceptable. Anybody know of any other obtainable telecentric objectives with better FF coverage?

If you keep an eye on ebay the 3x/5x/10x nikon objectives pop up for $200ish pretty often. I've seen them called toolmakers, measuring, and measurescope objectives, as well as EDF 200XX. The 20x and 50x i've never seen for sale used and I would expect them to be pricey from nikon.

I actually started doing macro with an mp-e and sold it to get those objectives. I don't know if it was my copy but the mp-e softened a fair bit past 3x. With a huge working distance and no additional optics needed the nikons were a nice introduction to working with objectives. Of course, now i'm scrounging surplus equipment for lenses thanks to this forum

I still use them whenever I stack and stitch though, which is fairly often. I plan on eventually switching to a medium frame camera, but until then I will continue to use them for all my sns work.

[rjlittlefield]

It's my understanding that the "correct geometric distortion" function in photoshop is meant to fix this kind of problem as they are coming with normal panoramas too. Am I wrong?

Yes, that's wrong. Photoshop can correct for the kinds of distortion that result from which direction you have the camera pointed, if the camera (entrance pupil) stays in the same location. It can also correct for shooting with the camera in different locations, if the subject is planar. The corrections in all these cases can be exact. But Photoshop cannot completely correct for distortions of a curved surface that are caused by changing the location of the camera. I don't have time to explain why that is, but there's an illustration of it two paragraphs down from here.

Parallax is only a problem when you have foreground and background objects. You might have some problems with some of the legs or the head, but the majority if your subject is a continuous, smooth surface.

Sorry to say, but that's wrong also.

To see the problem, consider a simple cylinder with its axis horizontal. A ring is drawn on the cylinder, going directly around it. If you position a camera with an ordinary lens straight out from the ring, then in the picture the ring will appear to be a straight line. But if you position that ordinary lens to the right of the ring and shoot a picture, then the ring will be seen from the side and in the picture it will be curved, concave to the right. If you position the lens to the left of the ring and shoot a picture, then the ring will be seen from the other side and in the picture it will be curved, concave to the left. If you try to stitch any two of these pictures together, there's no way to make the ring line up.

Straight, curved left, and curved right are just fundamentally different, and what creates that difference is the curvature of the surface combined with the movement of the entrance pupil. The differences in shape, and the resulting impossibility of lining them up, are a form of parallax error.

If you repeat this experiment with a telecentric lens, then in all three pictures the ring will appear straight. This is because the telecentric lens essentially looks straight forward from every point on its front element. As you move the camera sideways, the ring on the cylinder gets seen by different parts of the lens, in each case whatever part of the lens happens to be straight out from the ring.

I was under the impression that a normal 10x NA0.3 was close enough to telecentric as to not make too much of a difference.

That is true IF you simultaneously turn off alignment in the software so as to impose orthographic perspective, AND your setup is stable enough that you don't get misalignment artifacts when you do that.

It is the same for other high power large NA objectives. In addition some but not all low power objectives are close to telecentric. The Mitutotoyo M Plan Apo 5X NA 0.14 is close to perfectly telecentric. But other objectives such as the Nikon CFI BE 4X are very far from telecentric. I've measured the entrance pupil of the CFI BE 4x as being only 42 mm from subject [REF].

I do believe I may have rotated the camera between imaging the legs and the body, but I can't remember now (and I don't have access to the original images at this moment as they are on my other computer). Its not something I usually do but it does save time as it increases overlap in the limbs as they are essentially "portrait" in shape.

Well, when I look at just the top panel in http://www.photomacrography.net/forum/u ... 4239_1.jpg, the one that shows the head, I see that the left side of the panel is rotated about 4 degrees counterclockwise from vertical, while the right side of apparently that same panel is rotated about 4 degrees the other way, clockwise. At the same time the top of that panel has been bent outward, like barrel distortion. I would guess that the bottom of that panel is bent inward, like pincushion distortion, which is why I said earlier that it looks like the tiles have been distorted into keystone shapes.

It is true that parallax is often a problem with normal panoramas as well, but photographers get around that by panning the camera around the nodal point with a special panorama rig.

To be clear, this special "nodal point" is in fact the entrance pupil of the lens. See Theory of the “No-Parallax” Point in Panorama Photography for more discussion.

What telecentric optics do is to move that special point out to infinity, after which there's no difference between rotating around the entrance pupil and simply shifting the camera along the X and Y axes while keeping it pointed in the same direction.

Essentially telecentric optics provide the perspective of shooting from a very long distance with a very high power telescopic lens, but in a small package that works at low magnification.

--Rik

[cube-tube]

Thanks, Rik. I had never thought about that kind of parallax.

About setting up your stacking software to create a telecentric image: That should work with most objectives, including objectives that are fairly far from telecentric, correct? I am still not totally clear on that, because everything from foreground to background should be reproduced at the same magnification, but there could still be large foreground objects blocking portions of the background that would be visible (unobstructed) to a telecentric lens, so it doesn't seem like true telecentricity in that way.

[klevin]

Edmunds Optics has quite a few telecentric lenses, none of them cheap, and with mounts that are not designed for our "normal" photography. That said, it's a good place to look around. Their print catalog is good, too.