Mitutoyo 5x on Mitutoyo FS-60 quick test

[enricosavazzi]

Continuing my tests of the the FS-60 scope, now it is the turn of the Mitutoyo M Plan Apo 5x. Zoom is set at 1x, so the total magnification is 5x. Nikon D200, diffused flash, PMax (which worked better than DMap for this test image), 112 images at 28 micron slices, no post-processing of any kind.


Full image, reduced


center detail 1:1 pixel crop


corner detail 1:1 pixel crop

Compared to the 2x, corner quality is much more consistent with the center. Resolution is quite comparable to the 2x at the center, or slightly better. Defective pixel trains are much longer than with the 2x (which might mean the 5x is less telecentric, or that its objective mount is slightly decentered).

[enricosavazzi]

And this is with the zoom set at 1.5x, so the total magnification is 7.5x. 330 shots at 10 micron slices. Perhaps a bit too thin slices, but I run into possible banding problems with the 2x objective at 1.5x zoom and did not want to have the same trouble.


whole image, reduced


center detail, 1:1 pixel crop


corner detail, 1:1 pixel crop

Resolution is still good, definitely there is no need for an expensive and difficult to find Mitutoyo M Plan Apo 7.5x. In this image there are a lot of edge trails at the bottom produced by the Zerene auto alignment, probably because the average height (Z) of the subject was much higher along the bottom.

One more thing that I am seeing (unrelated to the performance of the Mitutoyo FS-60 and objectives) is that my D200 now has quite a lot of defective pixels, dozens across the whole frame. Illumination is by a powerful flash and ISO set at base, so this is not a high ISO or long-exposure artifact. Perhaps the D200 is trying to tell me it is looking forward to a comfortable retirement?

[rjlittlefield]

In this image there are a lot of edge trails at the bottom produced by the Zerene auto alignment, probably because the average height (Z) of the subject was much higher along the bottom.

Usually vertical trails like this at top or bottom of image mean that the optical axis is tipped up or down with respect to the movement axis. They occur commonly with long setups that sag.

I can't recall what your setup looks like. Have you posted a picture of it?

--Rik

[enricosavazzi]

In this image there are a lot of edge trails at the bottom produced by the Zerene auto alignment, probably because the average height (Z) of the subject was much higher along the bottom.

Usually vertical trails like this at top or bottom of image mean that the optical axis is tipped up or down with respect to the movement axis. They occur commonly with long setups that sag.

I can't recall what your setup looks like. Have you posted a picture of it?

--Rik

Hi Rik,

my setup is at http://savazzi.freehostia.com/photograp ... _fs-60.htm , Figure 3 down the page. I got no edge trails with the zoom lens at 1x and the same objective, nor with the 2x objective and the zoom at 1x and 1.5x, so for the moment I believe this is subject-dependent, and specifically depends on the Z height of the subject around the edges of the frame (the alternative being that the tube lens becomes occasionally misaligned during zooming, which would not be a nice discovery).

If I have time I may do some tests with an inclined paper sheet that extends past the edges of the frame as subject, which should remove all doubts (if the edge trails follow the inclination of the subject, then we have proof).

Then again, whether the edge trails appear or not may also depend on the optical formula of the objective (e.g. telecentric or not, and the distance of the entrance pupil from the subject), so any negative test results (= no edge trails occur) may say nothing about different objectives.

I have a direct experience with edge trails caused by misalignment in the optical path. If edge trails are not perpendicular to the edge of the frame, then there is misalignment (or a weird rotation of parts of the optical assembly around its axis relative to the subject, not impossible if a focusing helicoid is part of the setup). If the edge trails are perpendicular to the edges, then all bets are still open - although a misaligment is unlikely to be exactly perpendicular to the edge, it is still possible.

[enricosavazzi]

As an experiment, I processed the same stack shot at 7.5x, in reversed order. The result was not quite as I expected.



I have a possible explanation for the edge trails along the left margin, though - with the stack in this order, the first images contain focused detail along the right margin (the antennae), and Zerene latches onto this detail, adds it along the right margin of the final image and aligns all subsequent stack images accordingly. I am not so sure why the trails are also present along the top margin and not the bottom. Perhaps it has to do with the fact that the antennae are also located near the bottom of the frame so the bottom becomes "fixed" and any alignment happens at the top.

Rik, did you ever see images with edge trails along three margins? Or is this prevented by the aligning algorithm?

Incidentally, this stack has quite a lot of depth, 3.3mm compared with a field of view of approximately 3.2 by 2.4mm, so the extent of the edge trails is not surprising. It would seem to imply that this objective definitely is not telecentric.

[Pau]

I'm not stacking expert at all, but I guess the L shapped dust trails may be due to subject movement during the shot sequence.

[enricosavazzi]

I'm not stacking expert at all, but I guess the L shapped dust trails may be due to subject movement during the shot sequence.

Wiggly dust/defective pixel trails can indeed be caused by subject movement (although a single accidental movement is more likely to cause either an interruption of the trail or an "echo" in the image if the movement exceeds the capabilities of the alignment algorithm). However, in this case I don't think so. I base my conclusion on the fact that the same stack, processed by Zerene in opposite orders, produces very different trail shapes: one with a slight bend in the middle, the other L-shaped with a sharp bend at a different position. I still believe that the relief topography and the non-telecentricity of the objective are combining together with the alignment algorithm to produce the irregularities of the trails.

[rjlittlefield]

Enrico, thanks for the additional information. This last result actually is what I would have expected, given the earlier results.

Let's see if I can explain how things work. Please bear with me...

To begin,

Zerene latches onto this detail

It is not good to imagine that Zerene "latches onto detail" when doing its alignment. Actually Zerene makes no specific attempt to find details and match them up.

Instead, it works by essentially overlaying one image on another and adjusting the offsets, rotation, and scale so as to maximize the correlation of luminance values at all pixel positions in the whole image. This process does give some extra weight to focused detail (because the pixel values are changing quickly there), but with shallow-DOF images the alignment is strongly influenced by the 99% of all pixels that are not focused.

This use of overall correlation is the reason that asymmetric compositions sometimes produce bogus alignment defects. As a dark-to-light transition goes out of focus, the blurred edge can appear to move across the image. When that happens, Zerene will pay attention to the blurred edge. If there's not enough other information in the image to dominate the alignment, then Zerene will keep the blurred edges lined up, and as a result will incorrectly infer that a shift has occurred.

Similar problems can occur with scaling, even when the subject and composition are symmetric. The fruit fly face that I use to talk about Zerene is a bit more "bug-eyed" than it should be because of this effect.

There is more discussion and illustration of this problem at http://www.photomacrography.net/forum/v ... 9878#79878.

The alignment process always starts at one end of the stack and proceeds to the other. The starting image establishes framing for the whole stack. All other images shift to match that, based on the chain of image-to-image alignments.

With default settings, Zerene aligns the two end images against each other to estimate scale change, then begins the stack at the end with narrower FOV. With most stacks, this results in subsequent images expanding to more than fill the frame, which results in no edge streaks at all. However, if the alignment process is led astray by the apparent shifting of blurred transitions, then it can happen that subsequent images get scaled or shifted so that they do not fill the frame. In that case the non-covered pixel positions get filled in by interpolation from the closest edge. It's that interpolation process that produces the streaks.

You asked "did you ever see images with edge trails along three margins?". Sure, I've even seen streaks on all four margins. The default behavior tries to avoid streaks by starting at what appears to be the narrow end. But if that decision turns out to have been wrong and actually it started at the wide end, then some images can shrink away from the frame on all edges. If those frames happen to contain well focused detail at their own edges, then the interpolation process can produce sharp streaks that propagate into the final result.

Sharp streaks that propagate into the output can also degrade the alignment process. That's another reason why the default is to minimize them by starting at the narrow end. My guess is that your two different orders produced different dust trails because the different streaks degraded the alignment in different ways.

Now, a bit more about the interpolation process that produces the streaks... Zerene's process is basically driven by pixel positions in the final framing. For each of those pixel positions, for each source image, Zerene needs to retrieve a pixel value from the source image. So, it computes a pixel position based on the alignment parameters and drops an interpolating kernel over the source image at that position. If this position ends up outside the source image, as it does around the edges of shifted or shrunk images, then the resulting pixel value is obtained by just grabbing the nearest edge pixel. Where there's a wide band of pixels that are outside the source image, the same edge pixel gets grabbed again and again. That's what produces the streaks.

You mentioned that "If edge trails are not perpendicular to the edge of the frame, then there is misalignment (or a weird rotation of parts of the optical assembly around its axis relative to the subject...)". Edge trails not perpendicular to the edge of the frame will occur specifically when the alignment process has inferred a rotation. Of course that can mean that either there really was rotation, or that the subject & composition was asymmetric in a way that made Zerene see a rotation due to blooming of bright/dark transitions in OOF areas.

Detecting telecentricity and dealing with lenses that are not telecentric is an interesting and challenging problem.

Because of the blooming problem, it's not really reliable to decide telecentricity based on a deep 3D subject. Instead, I think it works better to use a flat subject that has a lot of detail all over it, and look at the scale change as the whole frame just barely goes into and out of focus. This avoids misleading indications caused by blooming, giving instead an accurate reading of scale change within the in-focus slab, which is what really matters.

If a lens is close enough to telecentric (insignificant scale change within the in-focus slab), then the best approach is to turn off Scale at Options > Preferences > Alignment. This will impose perfect orthographic projection on the assembled stack, just what you would expect if the lens were perfectly telecentric and the computational alignment process was also perfectly accurate.

If a lens is not close enough to telecentric to get good results by turning off Scale adjustment, then at present all you can do is to turn on Scale adjustment and hope that it works well. It would be better if Zerene allowed you to force a specific scale change per image, but that option is still sitting on the to-do list.

--Rik

[rjlittlefield]

my setup is at http://savazzi.freehostia.com/photograp ... _fs-60.htm , Figure 3 down the page.

Thanks for the detailed description. I agree this rig is very unlikely to have actual misalignment, so edge streaks in the stacked result must be due to errors in the alignment computation as discussed above.

--Rik

[shrek]

It would be possible that the head of the microscope has a light play ,I have it a which has of the play

jp

[enricosavazzi]

Rik, thanks for all the information.

I decided to do a few more tests to clear up some of my ideas. They may be useful to others, and they are about the same objective, scope and subject already discussed in this thread, so I think I can just as well continue here, rather than create a new thread.

In these tests, the stack is processed top-to-bottom along the Z axis. This is the same stack at 7.5x discussed above, only the processing with Zerene changes.

To see what happens in practice, I switched off alignment, rotation and scale in the options (brightness is still on because I noticed slight fluctuations).


full image, reduced


center detail, 1:1 pixel crop

Quite bad actually, so I did not bother with corner detail (it is worse than center). All trails are gone, of course. There is no obvious misalignment of the optical path, which should be quite evident in these conditions, but there may be a slight random shifting/wiggling in the XY plane from image to image (perhaps play in the focusing rack? perhaps the subject being blown around a few microns by the flash blasts?). Quite evident color haloes. In any case, not acceptable.

Next is a test with alignment turned on (still no scale, no rotation):


full image, reduced


center detail, 1:1 pixel crop


corner detail, 1:1 pixel crop

Center detail is superb (probably the best seen so far in the tests of the FS-60), but corner detail is visibly lower (but better than the preceding test). The edge trails along the lower margin are back. Dust and defective pixel trails are also back, but somewhat less wiggly than before. Color haloes are still there.

Finally, a test with only scale on (no alignment, no rotation):


full image, reduced

The dust and defect trails are short, straight, and they all point toward the center of the frame.
EDIT: I remove the following sentence because wrong, as discussed by Rik: [This means there is no significant decentering of the optics, camera or focusing rack].
Edge trails are absent. Resolution is poor like in the case of no alignment, no rotation, no scaling (small random wiggles everywhere), so I did not bother with detail pictures.

In conclusion, probably I could get away without rotation, but both scale and alignment are necessary to get an even resolution from center to corners. If I were interested in top center resolution at the expense of corner resolution, however, I could consider switching off both scale and rotation.

Rik, could the resampling necessary for scaling be responsible for the loss of fine resolution in the center areas?

[rjlittlefield]

Rik, could the resampling necessary for scaling be responsible for the loss of fine resolution in the center areas?

Sure. Whenever you have fractional pixel shifts, there will be some softening of pixel-level detail. The canonical example is that a half-pixel shift turns 0,254,0,254,0,254 into 127,127,127,127,127.

If preservation of that last bit of detail is critical, you might consider using Options > Preferences > Preprocessing > Pre-sizing to 200%. This should cause less degradation by the alignment method, at the cost of processing 4 times more pixels. In the canonical example, the pre-sizing turns 0,254,0,254 into 0,127,254,127,0,127,254,127, and the original half-pixel shift, now a full pixel, preserves the pattern as 127,254,127,0,127,254,127 etc.

However, in the specific case you're showing, I suspect the problem with central resolution is that you have a bit of lateral jitter in the focusing mechanism. When you don't allow to correct that jitter, and PMax combines the various jittered versions, resolution drops.

The dust and defect trails are short, straight, and they all point toward the center of the frame. This means there is no significant decentering of the optics, camera or focusing rack.

Not so. It just means you didn't allow for correction of any of those things, even if if they were present. When the only correction you're allowing is to scale, then the dust and defect trails have no choice but to be straight and radial, no matter what the optics are doing.

If you allow all corrections and still the trails are straight and radial, then you can reasonably infer that everything is stable and centered, and in addition you had a subject that did not induce alignment errors.

--Rik

[enricosavazzi]

The dust and defect trails are short, straight, and they all point toward the center of the frame. This means there is no significant decentering of the optics, camera or focusing rack.

Not so. It just means you didn't allow for correction of any of those things, even if if they were present. When the only correction you're allowing is to scale, then the dust and defect trails have no choice but to be straight and radial, no matter what the optics are doing.
...--Rik

True, wrong thinking on my part, so I am editing out that sentence.