Second attempt Nikon 10x objective - ant portrait

[lauriek]

This time a fairly small ant - entire ant is approx 3-4mm long, caught on the front wall of my house and frozen...

Shot with Olympus E330, OM Auto bellows at a bit less than 1/2 extension with the Nikon 10x CF plan objective. Macro twinflash for lighting.

Stack of around 60 images combined in CZM, with 33 as the number in find-detail.



I've got images for a deeper stack but I'm not sure if it will turn out any better than this. It's got a few minor halo effects but it came out of CZM pretty clean - just some minor CA removal in Photoshop, around some of the front facing facial hairs, and a tad of noise removal with Neat-Image. Probably neither show up at web size...

There's some minor ghosting on the front of the eyes from the oof antennae, but I'm not sure I can do much about that, although any ideas gratefully received!!

Here's a small crop at 100% pixel size for reference;



Comments appreciated!

[beetleman]

WOW comes to mind. Sure is a hairy fella.

[acerola]

One of my favorites are the ants. You made me want to crate more ant stack. This one is really good looking. The colors and details are perfect, maybe a little soft. The ant is maybe some kind of Lasius.

[rjlittlefield]

Yes, lovely!

You're correct about the light halo around the antennae, showing on the dark eyes. At this time, I'm not aware of any practical solution to that problem for this type of subject. Alan Hadley and I worked through some of the math a couple of years ago. It's theoretically solvable by subtracting out the component contributed by blurred foreground, but to make that work, you need several pieces of information that are not easy to get. Perhaps the worst is that you need a good model of the 3D point spread function of the lens. That varies a lot from between lens designs -- it's essentially what people are talking about when they say "bokeh". The problem gets even more interesting at high magnifications due to diffraction effects like Airy rings. There are some high end microscopy packages (e.g. from Vaytek) that handle this problem for transparent subjects, but I'm not aware of any that address opaque subjects like this ant.

I think I'm seeing a bit of blurring and doubling on the lower mandible. What does the full-resolution image show? There are also some swirly effects around hairs on the side of the head, typical of what Do Stack does with overlapping bristles. You might try Do Weighted Average if you're curious and have time. And of course, it would be interesting to see what TuFuse does with this stack.

--Rik

[lauriek]

Thanks guys!

Rik, yes there is a little blurring at the back of the lower mandible, there was some underneath as well which I cloned out already. At full resolution if you look hard it's far from perfect, but it's one of the best overall I've had so far at this kind of magnification...

Here's a 100% pixel crop of the area in question -



I should clarify when I say second attempt I mean second attempt with results I feel comfortable posting! I've done about 6 stacks with this ant under the Nikon objective and this is the only one I'm remotely happy with so far! (Plus gawd knows how many of those flowers the other day!) I think I need to spend a little more time getting used to the nuances of the software and a little less time shooting the stacks!

Looking forward to trying tufuse!! In preparation I just used CZM to align this stack and saved the aligned rectangles so will give it a go later on.

I think on the eye halos I will do some careful selected desaturation and maybe sharpening and see what I can come up with!

Péter, you know I thought of you when this popped out of CZM looking so nice!

[rjlittlefield]

Laurie, thanks for the blowup. This makes clear that what I was seeing as "doubled" in the first image is actually just bright reflections from the left and right sides of the same thick bristle.

Areas like this are tough for Do Stack to handle because the depthmap has to slew around from bristle to mandible to bristle to mandible to... I would expect better results from Do Weighted Average and TuFuse in this area.

About the eye halos, you might try playing with a levels adjustment, using a mask that's painted so as to apply only to the eye, with a gradual blend on the eye side. That's roughly equivalent to subtracting out the blurred foreground component, as I wrote earlier. You might need to do a little sharpening too, since having part of the aperture blocked by the antenna did cut into your resolution, same as stopping down would have. (One minor difference is that the blurring is asymmetric in this case -- like stopping down across the antenna, but not along it).

--Rik

[Charles Krebs]

Laurie,

Excellent. Really!

Sometimes when you're working with new techniques it's so easy to get caught up in looking for any problems and ways to make things better that, as the old saying goes, ... "you can't see the forest for the trees".

When I just sit back and look at the image... the reaction is like Doug's...WOW!

It would seem (based roughly on your info) that the width of this subject's head is just under 1mm wide. That would make the 100% crop shown about .25mm (.01 inch) wide.

So by all means take some time analyzing the results to learn what might be done in the future, but also have a great deal of satisfaction in a great image that would have been nearly impossible to do not that many years ago.

[augusthouse]

laurie, just out of interest, i took your image into Photoshop and raised the contrast significantly by 65+. the image can handle it.

[lauriek]

Charlie, if you and Rik like it that's good enough for me!

I concede I was very pleased when I saw this pop out of CZM and was only mildly disapointed when I looked at 1:1 and saw about 4 hours worth of careful cloning to make a /perfect/ image!

Thanks for the comments everyone!

I'll take a look at the contrast - I already boosted it a bit in RAW processing but it's always good to know you can push these image processing techniques further. I always tend to use absolutely minimal sharpening as well, I suspect I could push that further on the full size image. I'll maybe post back here when I get around to playing more with this stack!

I ran my first test stack with tufuse last night, and was quite impressed with the output - it's quite noisy in the oof areas but I can fix that easy enough, either by batch processing the whole input stack in neat image, or just cloning in a single oof background slice, will try to post results back here later on this evening or tomorrow if I don't make it tonight!!

ETA I just looked at what I did and I accidentally ran my first tufuse stack on a set of unaligned images, rather than the aligned set I'd prepared! Oops! That probably explains the odd noisy effects, just rerunning it now on the aligned images. It did an amazingly good job considering!!

[rjlittlefield]

I seldom do much with contrast on my images, other than level-adjust to compensate for deliberate underexposure to avoid blowing any whites. (I find that it's difficult to identify exactly which areas might blow, and I get really annoyed to shoot a stack, process it, and then discover that I'm missing detail in some bright area I didn't notice.)

On the other hand, I do sharpen heavily on almost all of my high magnification images. I figure it's only fair -- the tail end of the MTF curve always falls off pretty badly from optical effects, so I don't feel bad about pushing it back up with digital filtering. Since I'm a detail junkie, I'm always looking to make as much of it visible as possible. Even now, I'm often surprised by how much more detail shows up by aggressive sharpening.

Regarding alignment, I find that the DOF of these high mag stacks is so shallow that there's no significant scale change through the thickness of the in-focus slab, hence no need to correct for scale. If your focus slide moves smoothly, there's probably no need for shift either, in which case the whole alignment issue is moot. As I mentioned to Graham, my spider pedipalp can be processed without alignment, and that one has significantly smaller aperture and less magnificationthan what you're getting with this 10X objective. Running alignment probably won't hurt, but I'll be surprised (and want to know more!) if it helps a great deal either.

I'll be interested to see what you mean by "odd noisy effects". It's certainly true that TuFuse (and all the other pyramid methods I've seen) have a tendency to enhance noise. But the treatment of random noise won't be affected much by alignment.

--Rik

[lauriek]

At the moment my focus slide is half smooth - using the vertical (now horizontal) part of my micromanipulator - the fine focus is smooth, but I can only do part of the stack at a time with that, every 10-20 shots depending on magnification I have to reset the fine focus and refind my focus point with the coarse focus, then start again. Those refocussings do muck up the alignment slightly at several intermittent points through the stack.

Most annoying! I'm working on that, just got myself a nice set of 3 "Aerotech" linear stages, for fibre optic alignment or something, they seem to have ludicrously fine adjustment, and are sturdy as heck!

[rjlittlefield]

When I just sit back and look at the image... the reaction is like Doug's...WOW!

I can't say it any better than Doug and Charlie did.

This really is an outstanding image.

The lighting is just diffuse enough to show off the surface texture and details, while still strongly directional to model the shape and cast some interesting shadows. Head slightly canted, looking to one side, prominent cheekbones, faint smile on pouted lips... OK, maybe I've gone a bit far with that last part, but this image does strike me as being in the tradition of a classic "glamour portrait".

More, please!

--Rik

[lauriek]

Oh great, I hope my Mrs doesn't read your post Rik, due to the amount of time spent shooting and stacking this subject so far, she already thinks I have a possibly unhealthy interest in this ant!!

But thanks for the compliment, it's much appreciated!

More, please!

I'll do my best!

[lauriek]

Okay I've finally done some comparisons in Tufuse on this input stack.

First I should note that my initial attempts with tufuse were problematic due to the programs tendancy to increase contrast and noise, these both somehow seem to multiply up from the original images, unlike CZM, which I've been mostly using up to now. Hence the input files for this stack were reprocessed from the RAW originals. Where-as previously I had actually processed each input image for roughly the output contrast I wanted, this time I only increased the contrast very slightly, so the input TIFFs look very low contrast. (and also very low noise)

Here's my current output from Tufuse from these input files:-



It's still slightly lower contrast than my original stack of this Ant in CZM, and I have increased it slightly, but I'm sure there's room to push it some more!

This crop of the left antennae/top of head intersection shows up quite well one area where tufuse is worse than CZM (at least with minimal settings tweaks) and one area where it's better:-


(Top: CZM, Bottom: Tufuse)

Note the way Tufuse shows the top of the head line through the antenna. This obviously shouldn't show through but I have /no/ idea if there are any settings which might prevent this. This is a minus for tufuse, however in this instance that error would be easily cloned out.

Where Tufuse seems to win is the area to the left of that, which CZM turned to mush with confused halos. Tufuse did a brilliant job here, which to me makes it worth registering, as I wouldn't know how to begin using clone to tidy this area up, and trying to pull in tiny pieces of the originals by cross cloning would be a nightmare as well.

The only other area tufuse was worse is I think a result of it's current tendency to increase contrast, on the right side of the face the 'glow' from the overexposure it's produced on the head is somehow showing through the antenna.

So a couple of minor problems, it would certainly be appreciated here if the contrast 'enhancement' could be minimised. I really would prefer to avoid creating multiple sets of input images, not because of the time taken but the disk space overhead, I'm already accumulating images at a really scary rate!!

BUT from my own point of view, on this image at least, the problems in the Tufuse output are more easily corrected than the problems in the CZM output.

Incidentally I registered Tufuse pro! Couldn't be bothered with that command line stuff, I kept mucking it up!!

[rjlittlefield]

Laurie,

Great comparison -- thanks for posting.

The problem with eye showing through antenna would also appear (though perhaps less prominently) with Helicon Focus Method A and CombineZM Do Weighted Average. It is caused by the software having no concept of a foreground that should be rendered as opaque even though having similar or even less contrast than background behind it. That has been a problem with stacking software from the very beginning. I expect most cases of this problem will get knocked off soon, certainly within the next year.

In the meantime, there's a lot to be said for rendering with two different algorithms, then manually merging together the best parts of each. That's one big advantage of working with explicitly aligned images -- the processed composites can be lined up perfectly with each other. If you let each application do its own alignment, the composites usually do not line up perfectly. It's actually kind of amusing to make a Photoshop stack of the outputs from several packages and flash between them to see what sort of morphing got done by the various alignment methods. Sometimes the effect reminds me of claymation.

--Rik