6

This question already has an answer here:

I have a 500mm f/8 mirror lens, which is super cheap and light, but which offers terrible resolving power (and a concomitant difficulty in nailing its manual focus). This 50% crop is pretty much as sharp as it can get:

Sample crop from 500mm mirror lens

But one feature of the body I'm using is the ability to capture 12fps.

I've seen focus stacking done (usually with some difficulty), and I've also seen software (and now even some cameras) that produce "super-resolution" by stacking rapid exposures of the same scene. I suspect that, at least in theory, one should be able to produce higher resolving power or (equivalently?) lower Gaussian blurring resulting from an inferior optical system like this by incorporating the information in multiple samples of the same scene. Is that correct in theory?

In practice are there any methods or packages that can produce a sharper image from multiple exposures of nearly the same scene?

(To clarify: I imagine that one must stipulate that some minimal change in the scene exist between each image in the stack. But I also imagine that for "de-blurring" this would be on the order of the radial standard deviation that one is attempting decrease. E.g., for distant scenes and/or long focal lengths such small perturbations could conceivably be produced by things like atmospheric disturbance or shutter shake.)

marked as duplicate by mattdm, Hueco, xiota, inkista, Tetsujin Jul 27 at 11:54

This question has been asked before and already has an answer. If those answers do not fully address your question, please ask a new question.

  • Given the drop and the distance, was the focus already at infinity? If so, two things to note. Some lens may focus past infinity, so backing off just a hair might be sharper. If it's not having a focus past infinity, this might be just the limit of the mirror lens. I'm not sure if you could recover sharpness, but I have use image-averaging to lower the noise of a photograph shot at ISO 3200 with a micro four-thirds. – Calyth Apr 30 '18 at 19:54
  • @Calyth – I'm admitting a slight "focus miss" in the sample photo, but for this question I'm asserting that this level of sharpness represents the limit of the lens. (Surprisingly, even at the distance of this subject it seems to be inside of the lens's hyperfocal distance, so I am still manually adjusting for focus!) – feetwet Apr 30 '18 at 19:59
  • @mattdm: that sounds like the same question, except that in the end the asker was looking for (and accepted) the answer "superresolution," and that is explicitly not what I am looking for here (although I note that the idea is analogous, and I realize that the answer here might be encompassed in "superresolution" theory and techniques). – feetwet Apr 30 '18 at 22:12
  • To clarify, you're asking about a method to approach the theoretical resolution of the sensor with inferior optics, rather than to exceed it using superior optics as in superresolution, right? (I don't know how this could work, personally). – junkyardsparkle May 1 '18 at 1:27
  • 1
    @philberndt Yes, but that's for precisely the case that we're not talking about here: sharp images where the sensor resolution is the limiting factor. :) – junkyardsparkle May 1 '18 at 20:07
2

In focus stacking, one tries to only use the in-focus areas of every picture to generate a larger depth of field. Think of it as multiple pictures above each other with masks, and only the in-focus areas are opaque.

When referring to super-resolution, we usually talk about eliminating noise or doing very slight sensor shifts, therefore getting a higher SNR and/or higher pixel count.

Both techniques assume that one starts with decent pictures. Technology cannot make up information that simply is not there. Super-resolution needs spot-on photos to work, as does focus stacking. Having 200 shots of the milky way that are all way out of focus will not help you to get one very clear one, as the absence of information between the different shots will always be the same (circa).

Another example to get the idea of what is possible would be exposure stacking / HDR - you can get an HDRI out of an underexposed + a properly exposed + an overexposed picture, but take three equally overexposed pictures, and your "HDRI" would hardly differ from one of the original pictures. It is the same with focus.


Disclaimer: The following paragraph is only true if all your photos look like the one you have uploaded here. If focus varies, then technically, it would qualify as focus stacking to me.

The shown picture does not offer much in terms of differentiation e.g. between feathers - and since all of the photos will most probably lack this because it is not the noise of the sensor, but the lens's lack of resolution/sharpness, no software can accurately add this information. In theory, it would be possible to create a software that would add some feather-like texture to any out-of-focus bird, but I would not think that many people would be interested in it, as results would be horrible anyways. You might be able to get the eye a tad sharper if the bird moved around a bit, but that would count as focus-stacking (IMHO). If the main issue is resolution/sharpness of the lens, and not the focus (as it is in the sample image), then you could, in theory, get better results when panning/tilting the lens a bit, as it perhaps has a sweet spot for sharpness and you could get some details there. Again, this does not differ much from doing focus-stacking.

  • 4
    The problem with your "technology cannot make up information that simply is not there" is that it's not clear whether the information is or isn't there. If your 200 shots of the Milky Way are each different and only a bit out of focus, then perhaps statistical methods can be used to determine the most likely value for a given pixel in the final image. We're only looking at one sample image here, and it seems sharper in some spots than in others. – Caleb Apr 30 '18 at 20:15
  • @Caleb if all photos are equally un-focussed, then there's little hope (unless the lens offers no/little/bulls-eye bokeh). If the focus varies, then it qualifies as focus stacking. Will add a disclaimer about "with current information". ;-) – flolilo Apr 30 '18 at 20:19
  • 2
    I think that "no software can accurately add this information" overstates the case. In theory with a good enough model of the transfer function it should be possible to deconvolve it. The difficulty is measuring the transfer function. – Peter Taylor Apr 30 '18 at 21:11
  • @PeterTaylor How do you make information out of no information? I.e. if all information is the same, how can you inter- or extrapolate something? There's no way maths can do this, and as a rule of thumb: If math fails, everything else fails, too. (Of course, there might be slight deviations that could help with "restoring" some information.) – flolilo Apr 30 '18 at 21:55
  • Implicit in my question is the assumption that each image for the "stack" has some variation. I.e., it's obviously not n copies of the same data. I imagine that one would stipulate that the movement between each image must be some small (fractional) multiple of the blur (perhaps radial standard deviation) that one is attempting decrease. (It might be possible to create that delta by perturbing focus. At 500mm, I imagine it might naturally come from random shifts produced even by shutter shake.) – feetwet Apr 30 '18 at 23:55
1

In your situation Focus stacking is not the answer simply because there's nothing sharp in your image. Best to polish it up a bit in post and call it done or purchase some better glass.

This is about all I can get out of it:

enter image description here

  • Interesting: What techniques or methods did you use to produce that result? – feetwet Apr 30 '18 at 22:09
  • I'm guessing dehaze and sharpening. – John May 1 '18 at 13:10
  • 1
    Sharpness is contrast dependent, so contrast was adjusted so the darks of this image were shadows and the lights were highlights. Then smart sharpen with about 500, 10, 10 settings then a couple of rounds of finer detail enhancement using a Median layer. Still looks bad but it's an improvement. – R Hall May 1 '18 at 18:35
1

Super-resolution methods can in principle be used to get to higher resolution images. But the problem with the picture is that this is already unsharp while when using super-resolution methods, unsharpness will only appear at the desired higher resolution. Super-resolution and methods to improve sharpness ranging from simple contrast enhancing methods to sophisticated deconvolution methods, are used together to get to the desired final result. You then need many images not just to get to the necessary higher resolution but also to average out the noise.

Note that there is always a trade-off between noise and resolution. Given a single image with some noise, you can always reduce the noise by reducing the resolution by averaging over neighboring pixels. Similarly, given a large aligned image stack, you can average over the entire stack to reduce the noise, or you can decide to enhance the resolution to some degree by taking advantage of the shifts in the original unaligned images, but this then comes at the expense of the ability to reduce the noise by averaging.

In your case, it seems best to use the stack to only do noise reduction. The sharpened image in R Hall's answer looks quite good, but if you zoom in you'll see that the noise is also enhanced, which is quite typical when applying sharpening. A stack of images can thus be used to reduce this noise (you then need to sharpen each individual image and then average out the enhanced noise over the set of sharpened images). If the original image had been sharper, then with less sharpening needed the enhancement of the noise due to sharpening would be less, and that would give you more room to use super-resolution methods to get to higher resolution images without ending up with ugly noisy pictures full of artifacts.

0

Stacking to reduce noise and improve OVERALL sharpness is a technique which is already well established in astrophotography (it's called Lucky imaging) and Hasselblad have a proprietary version of it with one of their eyewateringly expensive MF digital cameras but surprisingly, nobody else to my knowledge, not Adobe or any other professional photography software product not intended for astrophography has implemented a purpose made version of this for the general photographer. The image you show has many aberrations (unsharpness, possibly diffraction, noise, poor contrast, poor bokeh). The other posts are correct that if all of your initial shots are not sharp, you will not get a sharp final image - you may improve it by Lucky imaging because you will be correcting other aberrations (mainly noise) and maximising what sharpness you can get. You can also get away with more contrast adjustment in post (and reduce the poor contrast) because you will have less noise. This technique however depends on a non moving object. It is difficult with a long lens, even with a good tripod and impossible where there is wind and moving living creatures: even with a high frame rates, the relative motion of twigs in that tree and the bird will create the equivalent of motion blur which is not correctable even with image alignment.

Not the answer you're looking for? Browse other questions tagged or ask your own question.