Interesting points to take home and ponder:
*1) The D800 with the 50/1.4G lens at F16 gives a lot more real end result image detail (as per MTF50 as well as in real life) than the D700 with the 50/1.4G lens at F5.6. The D800+F16 will take more sharpening, but it can handle it - a lot better than the F5.6 shot on the D700.
*2) The absolute corner of 50/1.4G lens at F11 is sharper than the center of the D700 used at the same settings.
So, yet another solid and physical proof of what some people (that HAVE tried and tested it in real shots) already knew, but general "internet wisdom" denies - If you want more DoF without sacrificing sharpness, and you have enough light to go for it with smaller apertures - get a camera with more resolution.
Thanks for the link. I don't even bother responding these days to "more MP means diffraction will be worse" posts.
One question for you since I suspect you've thought about it more than me and probably practiced it a lot more as well...
My assumption has always been that "stopping down to far" and then sharpening in post processing should be reasonably effective for IQ. You'll have a rather well behaved Airy disc to deal with in the center while on the edges things like coma and astigmatism which are very difficult to deal with in sharpening algorithms will be improved. This is of course apart from the primary issue of increasing depth of field.
Any thoughts or experience as relates to my assumption?
This long thread demonstrated to me that it was possible to recover quite a bit of resolution from diffraction smeared images using careful sharpening.
I think the bigger conclusion from this set of examples isn't simply "diffraction doesn't matter (much for typical shooting situations)," so much as "lens resolution doesn't matter (much for typical shooting situations)". The MTF plots show a 20-30% drop in resolution from peak to f16 (which doesn't exactly ruin the final images). Lots of people (myself included) go to extreme lengths to get lenses with <20% resolving power advantage over much cheaper, commoner alternatives --- but that difference doesn't create fantastic images that are clearly superior to what a better photographer could get with a kit lens.
In my own shooting experience, the downside to diffraction from stopping down too much is not so much the loss of fine detail (partially fixable with a little extra sharpening), but the drop in larger-scale contrast and "starburst" rays around bright spots. An image with high dynamic range or strong specular highlights (unlike the rather flat lighting of Roger's example photos) will suffer significant (and hard to fix) loss in "punch" as the larger-size-scale portions of the diffraction point spread reach visible levels.
mpmendenhall wrote:
In my own shooting experience, the downside to diffraction from stopping down too much is not so much the loss of fine detail (partially fixable with a little extra sharpening), but the drop in larger-scale contrast and "starburst" rays around bright spots. An image with high dynamic range or strong specular highlights (unlike the rather flat lighting of Roger's example photos) will suffer significant (and hard to fix) loss in "punch" as the larger-size-scale portions of the diffraction point spread reach visible levels.
Isn't that drop in large-scale contrast also pretty easily fixable in post-processing?
alwang wrote:
Isn't that drop in large-scale contrast also pretty easily fixable in post-processing?
Not really (in my experience). It's not simply a general global contrast reduction (which could be fixed just by "stretching" the histogram back out), but local changes across high-contrast edges and around bright points (it's only "large-scale" compared to micro-contrast/resolution/sharpness issues). Sure, you can crank up the "definition" slider (or equivalent), but the results tend to look pretty fake and "overcooked." Fixing washed-out streaks bleeding across, e.g., tree branches against a bright sky, or mushy haze around specular highlights, isn't going to happen in a visually convincing manner.
mpmendenhall wrote:
Not really (in my experience). It's not simply a general global contrast reduction (which could be fixed just by "stretching" the histogram back out), but local changes across high-contrast edges and around bright points (it's only "large-scale" compared to micro-contrast/resolution/sharpness issues). Sure, you can crank up the "definition" slider (or equivalent), but the results tend to look pretty fake and "overcooked." Fixing washed-out streaks bleeding across, e.g., tree branches against a bright sky, or mushy haze around specular highlights, isn't going to happen in a visually convincing manner.
I might be wrong, but I think using AEB would help some of the bleeding since you said it's more pronounced in a high DR scene.
hiepphotog wrote:
I might be wrong, but I think using AEB would help some of the bleeding since you said it's more pronounced in a high DR scene.
I agree, you might be wrong.
Diffraction effects an image by spreading out what would (in the impossible case of no diffraction whatsoever) be each sharp point of light in an image into a wider blob. Part of this is broadening the sharp (smaller than one pixel) central peak to a slightly wider bump (leaking over into a few adjacent pixels, causing reduction of fine sharpness). However, the diffraction point spread also has long "tails" around the central spot (the exact shape of which depend on the aperture shape; you can see this in the "starburst" pattern around bright points), and, the more you stop down, the brighter these tails get relative to the central spot. It's not a problem that can be fixed at all by changing the exposure: that'll just give you brighter or darker versions of the same washed-out image, with the darker regions equally "polluted" by tails of light from the bright spots. Exposure changes will only vary whether/how much you blow out the highlights. This problem is worse in higher DR scenes, because high DR is defined by the bright spots being a lot brighter than the dark spots, so a fractional percent of the light from bright areas bleeding into dark areas will be more visible.
mpmendenhall wrote:
However, the diffraction point spread also has long "tails" around the central spot (the exact shape of which depend on the aperture shape; you can see this in the "starburst" pattern around bright points), and, the more you stop down, the brighter these tails get relative to the central spot.
Just to make sure I'm following you in the details, the tails do not in fact get brighter relative to the central spot as you stop down if the aperture remains circular. All that would happen is their spatial extent would expand. The ratio of energy in the central Airy disk to that in outer rings (the tails) would remain constant. Yes?
What makes the tails stronger relative to the central Airy disk is that as we stop down the aperture is becoming less circular. More polygonal, number of sides depending on the iris construction. As this occurs more energy falls outside the central disk. Yes?
Yes, I think you've got the details right, so let me clarify my statement about "more outside the central spot."
If you keep the same shape aperture, then as you scale the aperture shape the diffraction pattern scales inversely. Thus, the statement "brighter tails relative to the central spot" depends on how you're defining what the "central spot" means. If you define the "central spot" to scale with the size of the diffraction pattern, then there's always a constant ratio of the amount in the center (e.g. the main peak of an Airy disk) to the tails. However, if you measure at a fixed radius (e.g., what fraction of the light spills outside a 20pxl radius, vs. how much is inside a few pixel region near the source), you get more of the light far from the source (bigger, more visible diffraction stars) as you stop down --- which causes the visual impact of diffraction beyond central-peak-broadening blur.
Changing from circular to hard-edged polygonal apertures further exacerbates the problem, by concentrating the diffracted light into easy-to-see (and hard-to-correct) spikes slashing across the dark parts of the image. If you had a perfectly circular aperture, unchanging across the frame, then this would be easier to fix with a simple mid-frequency contrast boost --- but accurately de-convolving a slightly irregular streaky star pattern (which changes across the frame) is a whole lot harder.
The reason I was thinking in terms of energy inside/outside the first null as an important figure of merit was that I'm assuming the central disk to be straightforward to deal with via sharpening/deconvolution using a reasonably sized and stable kernel. At least for any aperture likely to be used by a photographer.
Energy past the first null is going to be difficult to deal with and will of course drop local contrast to some degree. Whether that can be dealt with using local contrast enhancement techniques is another question. So in that sense things that throw more energy out of the central disk might be considered worse than just scaling the point spread function spatially.
Too bad lenses don't typically come with drop-in slots for fixed aperture stops. With a precision circle --- or, even better, a bit of apodization for a soft-edged aperture --- you should be able to make a lens much more "well behaved" at small apertures, so all the lower-frequency effects could be cleanly reversed in software (instead of leaving irregular streaky artifacts with overly touchy deconvolution kernels).
EDIT: this actually won't help for deconvolution in the worst cases (e.g. cleaning up around bright specular highlights), because if any of the bright spot is clipped beyond sensor DR then you've lost the information about how much bleed-over light to correct for. It still might make diffraction more visually pleasing (when you don't want the star effect), lower the power outside the central peak, and be nearly fully correctable whenever clipping is avoided.
I hope more and more cameras go the "apps" route because I've always wanted an apodization multiple exposure mode. I think some film camera in the past actually had such a thing - a Pentax or a Minolta maybe.
Well, the reality is I can throw as much firmware, software and optical perfection at my photos and most of the time they'll just suck because of me
Better than apodizing multiple exposures --- bring back leaf shutters and vary the aperture during a single (potentially short) exposure. That shouldn't be hard to do with modern electronic aperture control. A more rounded aperture would still be nice; perhaps you could put the iris unit on bearings and spin it around during exposure, or rotate it by one edge segment with a USM drive.
Of course, a plain drop-in slot probably won't break one month after the warranty expires; and, as a cross-camera-platform manual lens user, I prefer my lenses not to depend on whizzy electronic fanciness that won't be compatible with my next different brand of camera body.
Diffraction effects an image by spreading out what would (in the impossible case of no diffraction whatsoever) be each sharp point of light in an image into a wider blob. Part of this is broadening the sharp (smaller than one pixel) central peak to a slightly wider bump (leaking over into a few adjacent pixels, causing reduction of fine sharpness). However, the diffraction point spread also has long "tails" around the central spot (the exact shape of which depend on the aperture shape; you can see this in the "starburst" pattern around bright points), and, the more you stop down, the brighter these tails get relative to the central spot. It's not a problem that can be fixed at all by changing the exposure: that'll just give you brighter or darker versions of the same washed-out image, with the darker regions equally "polluted" by tails of light from the bright spots. Exposure changes will only vary whether/how much you blow out the highlights. This problem is worse in higher DR scenes, because high DR is defined by the bright spots being a lot brighter than the dark spots, so a fractional percent of the light from bright areas bleeding into dark areas will be more visible....Show more →
Definitely did not take into account of the equally "polluted" problem. I just did a quick test myself at lunch and indeed the light bleed problem remained the same between a single shot and a 3-shot HDR. In fact, the single shot looks a bit better .
A spinning iris! love it. You'd need a pretty high rotational rate for fast shutter speeds though - 1/500 = 30,000 RPM if you want it to make a full rotation during the frame. Oh, just one segment would be enough to smooth things out I guess, so divide that by ~10, not bad.
At slower shutter speeds, I could imagine a spinning iris mechanism that could a. produce almost any arbitrary (convex, at least) iris shape, and b. create variable apodization. Just imagine a single blade coming in from the edge, spinning. Vary the thickness v. radius to adjust apodization. Vary the length and/or thickness v. spin angle to adjust the aperture shape.
2 general notes since I dont have much to add to the very interesting conversation going on in regards to how aperture shape as one stops down effects imaging.
1. Zeiss figured out some time ago that additional resolving power in the sensor and lens did not act in a min(lens, sensor) way but rather that there is an increase in IQ with the same lens by going to a better sensor. (ie, another +1 to kwalsh)
2. this discussion reinforces the usefulness of movements as a practical means for increasing depth of field in an image beyond just increasing it around the plane of best focus.
kwalsh wrote:
I hope more and more cameras go the "apps" route because I've always wanted an apodization multiple exposure mode. I think some film camera in the past actually had such a thing - a Pentax or a Minolta maybe.
It was the Minolta Dynax 7: multiple exposures onto film with varying apertures. Minolta was always creative about focus and blur, from the 24mm VFC to their masterpiece 135 STF. Re Waterhouse stops, they did that, too, with the TC-1 über P&S. The 28mm lens alone, based on pics I have seen, is a wonder of OOF blur.
Yes, both enhanced resolution and a greater depth of field can be quite noticeably improved (in a from the start good optical construction) by changing the aperture shape.
The normal round or polygon shape makes diffraction rings that's actually visible in a normal good camera/lens combination. And the OOF areas get flat-top defocus point spread functions, making bokeh hard and creates spurious resolution doubling (when the image contrast inverts). Combining two images where the rings don't coincide increases sharpness by making the sharpening process a lot more predictable. This is one of the advantages with the 135STF - that is a remarkably high contrast lens within the entire depth of focus.
One problem is that any way to exploit this is going to be quite expensive. And the normal focal plane shutters makes aperture-varying during the exposure impossible.
And BTW-
That resolution was a combination of several parts in the system, and that improving one made the end result better even when the rest of the system stayed the same was known even in the early 20's. We didn't have a scientifically sound way to predict the end result given two or three system parts until the 40's though, when Schade, Selwyn and Luneberg started to home in on a workable model.
.