[microlight]

A great image and one that will be remembered as iconic, depends on many factors. Resolution will never be one of them

Agreed, Peter. I can't pretend that any of my images will ever be iconic, but I enjoy seeing what various techniques work (or don't work) with my camera.

[biz-engineer]

I understand there are occasions where super high resolution is required, but for 99% of the time for casual or serious ameteurs it is not...A great image and one that will be remembered as iconic, depends on many factors. Resolution will never be one of them

Taking iconic images is not my goal.

The reason why I'm still interesting in high definition files is because I do the full exposure to paper print process. Up to 20"x30" the print output quality is photographic quality.
At 30"x40" the even the 36Mpixels of the K1 don't render like a fine photograph anymore, the print output renders more like a commercial advertising poster, a little better than most poster, but definitely not photographic quality).

Another thing to consider: the price of photo prints doesn't depend on the quality of the files I send to the lab. Prices increase proportionally with the size of prints. So when I order four 30"x40" prints, iconic or not, I prefer to have high output quality for the same price.

---------- Post added 09-04-20 at 10:12 ----------

True world has a gradient from 100 to 0 value (with pure 100 to the left and pure 0 to the right). Gradient is one pixel pitch wide for simplicity.Images:75 - 25 (each pixel is averaging the left half and the right half of the gradient, the true gradient is perfectly centered on the pixels)90 - 40 (each pixel is averaging left and right halves but shifted less than a half pixel to the left a bit so it's more bright overall)50 - 0 (shifted a full half pixel to the right, so the left pixel is taking the center of the first image and the right has only pure black. This is the best sub-pixel-shift case.)Now you expand these images to 3 pixels wide with interpolation75 - 50 - 2590 - 60 - 4050 - 25 - 0Best alignment can be done to something like this:90 - 60 - 4075 - 50 - 25------50 - 25 - 0Stacking would output82.5 - 53.3 - 30 - 0or somesuch, that is a much better description of the "true state" of a linear gradient from 100->0. More interpolated pixels and more frames with different offsets would help fill this out and capture the full range.Of course it works well in this case because the linear interpolation is "correct" -- but it works even if this is not the case with enough frames, just not quite as quickly/obviously.

Thanks a lot for that great example, which helps a lot to understand the process.

---------- Post added 09-04-20 at 10:22 ----------

In my experience if your goal is to add detail in a landscape situation, you are far better shooting multiple images panorama style and stitching.

That's a good point. I learned by experience that panorama stitching using rectilinear perspective projection (to equate a single exposure) works best with distant subjects and/or with not too wide field of view.

In the case of W field of a field of view wider than ~30mm FF equivalent, the edges of images are stretched so that to fit a spherical surface for a rectilinear rendering, and at 20mm FF equ. the edge resolution of the stitched panorama becomes no better than resolution of a single exposure. There is also the problem of parallax error and precision for fitting of frames in presence of pronounced perspective distortion. So in the case of a wide angle scene, the pixel shift or super-resolution method has the advantage over the stitching method.

---------- Post added 09-04-20 at 10:29 ----------

In any case, If you fit the three highest values of the cross-correlation curve to a quadratic equation and then solve for the location of the peak of that curve, that location value will be a decent estimate of the sub-pixel offset.

I see. Image frames, even after best possible alignment, would produce a cross-correlation error, due to incremental information present in one file relative to the reference image. For the method to bring sub-pixel resolution requires significant up-sampling before the alignment process, something like 4x or 8x the original resolution, stacking and down-sizing 2x or 4x. Processing via some GPU / hardware acceleration wanted !

[photoptimist]

I see. Image frames, even after best possible alignment, would produce a cross-correlation error, due to incremental information present in one file relative to the reference image.

Exactly. If two frames happen to be offset by an exact integer number of pixels in both X and Y, the cross-correlation would be extremely close to 1. Only the sensor noise and photon shot noise would prevent the the aligned frames from being identical. If the offset has a fractional pixel component, then none the cross-correlation will be near one but their discrepancy from 1 will follow a mathematical pattern related to the fractional offset.

For the method to bring sub-pixel resolution requires significant up-sampling before the alignment process, something like 4x or 8x the original resolution, stacking and down-sizing 2x or 4x. Processing via some GPU / hardware acceleration wanted !

Estimating the subpixel alignment between frames does not require upsampling. It can be done at the native resolution of the camera. The shape of the cross-correlation surface sampled at integer offsets provides the data needed for estimating the subpixel offset between pairs of frames.

Nor does computing the superresolution image require massive upsampling of the raw images either. There are mathematical methods (Inverse problem - Wikipedia) that can directly estimate each super-resolution pixel value from the un-resampled raw images. In fact, for best results, the "correct" approach should not even demosaic the RAW image (demosaicing would actual cause more problems than it solves). Rather than demosaicing and upsampling all the camera data, the system computes the geometric coordinate relationships that say where each output superresolution pixel resides in each input camera data file. In essence, the inverse problem math basically unblends the superresolution pixels that were averaged together in the camera images.

That said, the math of inverse problems is computationally intense because it involves fun things like matrix inverses.

[weverka]

I understand there are occasions where super high resolution is required, but for 99% of the time for casual or serious ameteurs it is not. Look at the great photographers of the past working with the available technology, high resolution was a pipe dream.

A great image and one that will be remembered as iconic, depends on many factors. Resolution will never be one of them

I don't think you are right. low resolution pictures are fine for tv monitors, but fine art prints are done at high resolution.

As for photographers of the past, an 8x10 inch view camera, shooting at f40 gives a 25 micron resolution for 8,000 by 10,000 spots- 80 megapixels.

As a kid, I worked for a pro shooting 8x36 inch film in a panaram camera. At f16 we had a gigapixel.

[ThorSanchez]

I don't think you are right. low resolution pictures are fine for tv monitors, but fine art prints are done at high resolution.

As for photographers of the past, an 8x10 inch view camera, shooting at f40 gives a 25 micron resolution for 8,000 by 10,000 spots- 80 megapixels.

As a kid, I worked for a pro shooting 8x36 inch film in a panaram camera. At f16 we had a gigapixel.

He said that super high resolution isn't required for 99% of the images taken by casual or serious amateurs. I don't think much more than 1% of typical amateur work is high resolution fine art photography, although I guess I could be mistaken. My feeling is that most of us take a lot of cats and kids playing soccer and their families on vacation (back when travel was a thing).

[photoptimist]

He said that super high resolution isn't required for 99% of the images taken by casual or serious amateurs. I don't think much more than 1% of typical amateur work is high resolution fine art photography, although I guess I could be mistaken. My feeling is that most of us take a lot of cats and kids playing soccer and their families on vacation (back when travel was a thing).

That may be true.

Regardless of the 99% of snapshots that can be low resolution, I'd think that many "serious amateurs" aspire to be able create images that could fall into the 1% category.

It's like the cargo or towing capacity of pickup trucks. It goes unused for 99% of trips but people buy pickup trucks for the aspirational ability to handle the 1% scenario.

[ThorSanchez]

That may be true.

Regardless of the 99% of snapshots that can be low resolution, I'd think that many "serious amateurs" aspire to be able create images that could fall into the 1% category.

It's like the cargo or towing capacity of pickup trucks. It goes unused for 99% of trips but people buy pickup trucks for the aspirational ability to handle the 1% scenario.

I hate aspirational pickup trucks. I certainly agree that people often buy things to their 99% use case, when the 90% use case with work-arounds for the other 10% would save them many thousands of dollars.

I live in a part of Maryland that got literally zero snow this year and often gets a few inches a few times a winter. But many or even most people spec out their cars (sorry, full-sized trucks and SUVs) with full four-wheel-drive capabilities just for those couple of days a year. My wife drives a Jeep Wrangler that's regularly exercised to 25% of its capabilities.

---

Last edited by ThorSanchez; 04-15-2020 at 09:38 AM.

[house]

I hate aspirational pickup trucks. I certainly agree that people often buy things to their 99% use case, when the 90% use case with work-arounds for the other 10% would save them many thousands of dollars.

I live in a part of Maryland that got literally zero snow this year and often gets a few inches a few times a winter. But many or even most people spec out their cars (sorry, full-sized trucks and SUVs) with full four-wheel-drive capabilities just for those couple of days a year. My wife drives a Jeep Wrangler that's regularly exercised to 25% of its capabilities.

It's a shame this urge is played on so well in our societies. There's not much support to be had for making the adult decision to buy only what we can take advantage of. The childish urge to have everything is fed all the time.

[weverka]

He said that super high resolution isn't required for 99% of the images taken by casual or serious amateurs. I don't think much more than 1% of typical amateur work is high resolution fine art photography, although I guess I could be mistaken. My feeling is that most of us take a lot of cats and kids playing soccer and their families on vacation (back when travel was a thing).

he did say that. But that part of what he said is not what I was responding to.

He said "Look at the great photographers of the past working with the available technology, high resolution was a pipe dream."
That is just plain wrong. A 35mm film camera had the same resolution as today's full frame digital camera. And people shot plenty with larger format than that.

[ThorSanchez]

he did say that. But that part of what he said is not what I was responding to.

He said "Look at the great photographers of the past working with the available technology, high resolution was a pipe dream."
That is just plain wrong. A 35mm film camera had the same resolution as today's full frame digital camera. And people shot plenty with larger format than that.

I think that is at least debatable, and certainly not the consensus view. The consensus appears to be that 35mm slide film has the rough equivalent resolution of 6-20Mp digital depending on a number of factors. The K-1 is 36Mp.

[weverka]

I think that is at least debatable, and certainly not the consensus view. The consensus appears to be that 35mm slide film has the rough equivalent resolution of 6-20Mp digital depending on a number of factors. The K-1 is 36Mp.

You are right, the K-1 pixels are about 5 microns across, so you would need a fine grain film, that can make >100 lp/mm to be comparable.

[Rondec]

I guess the question I would have is which would give better results, shooting a bunch of extra hand held shots and combining them for super-resolution or running a single image through a program like Topaz's Gigapixel AI?

[texandrews]

I guess the question I would have is which would give better results, shooting a bunch of extra hand held shots and combining them for super-resolution or running a single image through a program like Topaz's Gigapixel AI?

I'd say multiple shots if you can manage it, based on what I've seen with G AI (which I own). G AI can be tricky to use, although the results can be very good. I use G AI with my Z, which of course doesn't have PS. Since I have a Z, the situations where I would use my K1mkII instead become less PS or *other* scenarios, so take that into account in my response...

[photoptimist]

I guess the question I would have is which would give better results, shooting a bunch of extra hand held shots and combining them for super-resolution or running a single image through a program like Topaz's Gigapixel AI?

That's a very interesting question.

The multiple-image super-resolution approach will collect and process the data to fill in the subject matter details that were actually in the scene.

Topaz's Gigapixel AI will fill in the details that the AI expects to be in the scene based on all the training data. That is, it fills in what is typically found in such scenes.

Thus, Gigapixel AI images will typically look right but the super-resolution image will actually be right.

That then gets into one's personal philosophy of photography. If photographs are supposed to show what is truly in the scene, then super-resolution is way to go. If photographs are only supposed to be pretty (not true-to-life), then Gigapixel AI works.

[texandrews]

That's a very interesting question.

The multiple-image super-resolution approach will collect and process the data to fill in the subject matter details that were actually in the scene.

Topaz's Gigapixel AI will fill in the details that the AI expects to be in the scene based on all the training data. That is, it fills in what is typically found in such scenes.

Thus, Gigapixel AI images will typically look right but the super-resolution image will actually be right.

That then gets into one's personal philosophy of photography. If photographs are supposed to show what is truly in the scene, then super-resolution is way to go. If photographs are only supposed to be pretty (not true-to-life), then Gigapixel AI works.

Normally I agree with you 100%. On this one I have to disagree somewhat. I did 2 shots for the Hirshhorn Museum and Sculpture Garden this fall. It required me to shoot a "scatter" of notes from Duchamp's Green and White boxes for our Levine collection exhibition (a big donation of Duchamp's work to us).


I shot on a copy stand in the Photo lab a setup prepared by the conservation department on top of a sheet of foam core. This was a little hairy, as the paper conservator and the curator had prepared this in the Conservation lab, and then rolled down the hall on a cart to the Photo lab, and none of the bits of paper, all facsimilies overseen by Duchamp himself in an extremely limited edition, were or could be anchored down. I was required to have my camera (a Z) already set up with the lights and essentially focused and exposure set with dummy papers I used, and then the foam core set up was carefully placed down on the bed ( a custom bed we had fabricated for us that rolls on the floor and is about 6 feet wide by 48-ish deep by maybe 18" tall). The foam core was around 60" wide. I had only several minutes to shoot, and only one chance---as we worked the individual papers began curling from atmospheric change---I used my own LED lights to cut down on the heat. I was not allowed to move the camera---so although I would like to have been able to stitch together 4 or 6 images, we couldn't do this ( I wasn't consulted on how to go about this, I wish I had been...)

So, because the aim was to produce a 1:1 ratio print to use as a base/background in a vitrine for the actual sketches and notes, and in the case of the White Box other ephemera and folders. As all of the originals are extremely light sensitive, they are constantly rotated, but the prints we had made stay in place as a referent for the whole contents of the boxes, if that makes sense. I can't share the photos I took because of permissioning ( The Levine's are still alive, so they technically still have ownership).

So, in steps G AI. You'll have to take my word for it, but it performed brilliantly and I wouldn't characterize the result as fake or "for pretty", as the Amish say. Viewers are able to scrutinize the originals against the print at 1:1, side by side. The results were so good, that we decided that we'd best do the prints monochrome so there was no confusion for the viewers.

As I said before, G AI is a tricky program to use---it took me a few tries to make it right (Hint: turn off "Auto" and dial in the 2 sliders manually...). But when it works, it works wonderfully. I'd also say that for a project like what I have described here, the Topaz claims of 6x uprez are outlandish---yet I was able to get where I needed to go with some room to spare. But this was a true torture test of the software I think. For most people's use, where the referent is not sitting right next to the print, I'd say a great deal can be done with this software by a sensitive user. And I'm sure we'll see improvements. I won't hesitate to use it with my personal work. There are lots of situations that don't lend themselves to multiple frame shoots.