Stacking 10 images in P'shop to raise rez

[greenmorpher]

Hello folks

Time to jolt everyone out of their holiday sloth!

The guy who posted this image in DPReview https://www.dpreview.com/forums/post/58963476 said he got this dramatic effect with a 10 image stack in Photoshop.

Presumably that means HDR. It sure looks like a strong HDR effect. For the subject, I like it.

But he is an m43 user and says he used the stack to raise his resolution from 16MPX to 60MPX so he could make a 20x30" print @ 300dpi.

Huh?

Does anyone know what he is talking about and if so, how to do that in PL? I don't mean how to achieve the HDR effect -- I do my HDR in Bracketeer -- but how to get that huge resolution increase.

As ever, I would be most grateful for elucidation.

Cheers, Geoff

[cathodeRay]

In a spirit of being un-slothful: I am not sure I do understand what he's getting at. I can't see how (vertical) stacking could ever increase (horizontal) pixel counts. Indeed, the pixel count is fixed at shutter/scan time and can only be changed by cropping (always down) or resizing (up or down). Maybe what he's doing is indeed HDR - would fit with multiple stacking perhaps? The resolution thing is something else (and usually, so far as printing is concerned, a simple finished image size by dpi calculation, and you either have the pixels or you don't...).

cR

[Hoogo]

Let's say you have some fine structure.
If your sensor (the grid) would be just that fine, you'd get a perfect picture.
Unfortunately the resolution of your sensor is too rough, only half of what you need.
4 Pixels hit one cell of your sensor, and the result is an average of them. You get no idea about the details when you look at the result.
But now you move your camera and take 3 more pictures: Half a sensor pixel right, half down, half into both direction.
All 4 results don't make much sense.
But then you average them all: Now you see details, somewhat blurred, but that can be handled by deconvolution.

Years ago there were special tools for this "superresolution". I don't know what PS can do for you today.
I remember that there was a camera that supported this sub pixel movement by design, Hasselblad?

With normal equipment I would take a few pictures with slightly tilted camera, scale the pics up by 400% and then place and mix them with some feel. Layers need opacitiy of 1, 1/2, 1/3, 1/4... , like in the attached PLD.

[bruce1951]

Many professional printers can 'rip' a file to get a larger than 'standard' resolution. But aren't you in reality simple 'inventing' detail that the camera never captured?
Besides, why would you bother apart from the exercise itself? I have had a 4000*3000 image printed, by a pro, 40"x30". Looks fine.
But I guess if you are going to crop heavily and zoom then just maybe the extra effort is worth it. Maybe!

regards

[bkh]

Many professional printers can 'rip' a file to get a larger than 'standard' resolution. But aren't you in reality simple 'inventing' detail that the camera never captured?

Yes, that's just upscaling (if your image has 100 ppi and the printer prints at 300 dpi, then that's what you need to do). But that's different from the super-resolution technique mentioned above.

Btw., if you aim at better image quality, then you are much better off taking a 3 x 3 panorama instead (if only because your lens only has to deliver half the resolution, and because the pixel area acts as a low pass filter). The necessary deconvolution also increases noise a lot.

if you look at the stacked image at 100%, then the quality isn't really convincing – I doubt that it has more detail than one of the original images.

Cheers

Burkhard.

P.S. If you want to experiment with super-resolution images, then have a look at Hugin. It's free and can align images at sub-pixel distances and merge them into higher resolution images. (Don't know if it can do the required deconvolution, though).

[bruce1951]

Thanks Burkhard. I guess my comments were more about how big/detailed do you go. Now days cameras and lens combinations can do a fantastic job. There's more chance of destroying an image in editing.

But I guess there are folks who always seek the limits just for the heck of it. Just Google some of the Gigapixel images out there. They are so large, and detailed, it make your brain hurt.

By the way. I have Hugin.

regards

[Hoogo]

Usually stitching and superresolution both don't work for moving objects, as it takes time to take all the pictures.
It should not work well for waves, so I guess only the beach and maybe the sky are truly stacked.

[photoken]

Yeah, stacking images doesn't really do squat in terms of increasing resolution/pixel count.

The only use of image stacking is in overcoming the limitations of the camera lens's depth of field -- you take multiple images, changing the point of focus between each. Then you combine the images using masking that eliminates the out-of-focus areas.

This is most useful in macrophotography, and there are any number of "focus stacking" programs to accomplish this.

[maxwell]

Hi,

as far as I understood the dpreview discussion, the 40 Megapixel (jpeg) respectively 64 Megapixel (raw) mode of the Olympus OM-D E-M5 Mark II is discussed, which works like Hoogo has explained (via sensor shift). In the description of the mode they speak of 8 pictures which are "stacked" to one picture. It would be nice if this mode of the new Olympus cameras can also be used with Photoline.

Cheers

Josef

[cathodeRay]

I can't help wondering if we are straying into the realms of voodoo here, or, perhaps rather more mundanely, marketing. A bit like the scanners sold these days as 7000+ dpi when there real world resolution is nearer half that figure. They may (or may not) collect say 7200dpi, but we only see at 3600dpi (the 7200 is sort of real, not interpolated, but it doesn't deliver real world 7200). I had a look at this when I was deciding what resolution to scan 35mm slides at, and came to a pragmatic decision that (shock horror) 2400dpi was OK for hand held holiday snaps (even if taken on what was a highish end SLR with a good lens in its day) that were never going to be enlarged above 8x10 (if that) or cropped, 3600dpi felt better for more special slides (though maybe not really getting much more detail) and by 4800dpi we have probably reached the point where we are getting all the data even a very good slide contains. Above that we get bigger files, but no more information (or maybe just resolve the grain - but that's really noise). I also came to appreciate that as great if not a the bigger challenge may well be not resolution, but dynamic range (eg as in that dog slide posted earlier).

We're also in 'pi' territory - dpi, lpi, ppi etc. What we are really interested in here is ability to discriminate between objects (hairs on someone's head, say, or grains of sand on a beach), or, under test circs, line pairs. I still stick to my earlier point: vertical stacking cannot increase horizontal resolution. Wearing a Newtonian rather than a quantum hat, no amount of extra info in one dimension can tell us anything about a second dimension at 90 degrees to that first dimension. I take Hoogo's point about sub-smallest unit (a pixel) offsets (which are of course in the second horizontal dimension) and then recombining as a theoretical proposition but I would like so see real proof of how this technique actually revealed detail at the sub-pixel level (because I struggle with how you can get real useful info at a sub-unit level). It may be this sort of thing is already being done, perhaps in a forensic, surveillance or medical setting, in a crude way like CT scanning reassembles a more complete image from many individual images. But again, I don't think you can see CT scanning in any way as vertical stacking, and of course CT scanning involves a lot of radial movement, in effect movement in the horizontal plane.

I'm probably getting it all completely wrong, but all in all a very interesting discussion!

cR

[Hoogo]

Found a nice example when I looked for that Olympus on dpreview. It also shows the problems of moving objects, and that that camera does a little more than just 4 pictures.

A little offtopic about stacking:
If I remember correctly, museums take many pictures of their art in different light or with different color filters. This way they capture more different colors than the usual 3 bands of RGB.

[bruce1951]

Technical advancement is all well and good. But at the end of the day the user, and as a result editor, will make or break the results.
A good example are cameras such as the Nikon D800 series. High resolution cameras. Many folks think that buying and using one of these cameras will give them great results. True. But unless used correctly they also offer a greater chance of poor results. ie the less room for error.
Likewise stacking software can produce excellent results. But only if the user has a good understanding of capturing the subject in the first place.

So my guess is that these new cameras are/maybe a technical break through that 'can' work but only if all the rules are followed. Otherwise it's just another marketing ploy.

As the manufacturers climb the technological pyramid the room for improvement get smaller and small just as the peak of the pyramid gets smaller. I also think that software developers are on a similar coarse. Consumers are simply looking for a technical answer to replace hard learned experience.
The effort put into improving what we already have would be a wiser investment than looking for the next new breakthrough.

Much of the world in now content with photos taken with a phone. The market for 'new' gimmicks is getting smaller and smaller.

IMHO

regards

[cathodeRay]

Thanks Hoogo, that's a good link, and I now get how it works, and furthermore can't disagree with the evidence in the OM crops of the tower - the level of detail is truly amazing (always was an OM man m'self, good to see they are still pulling the tricks). But, at the risk of being nit-picking, I still think the 'magic' is happening in the horizontal plane, by using the half pixel offset in that plane, and that talk of stacking, which by it's nature happens in the vertical pane, confuses the issue. I also note that the magic happens in the camera, and so again find myself struggling to see how stacking in a layer capable editor would be able to achieve the same aim. The pixels in the editor must surely be locked to their registered/in their relative position - not sure how I can see a way of somehow injecting alternate pixels to do the (horizontal) expansion - though of course the camera does somehow do just that... so somehow it is possible.

cR

PS Bruce - I see you are also inclined to see the thing as a bit of a marketing ploy - which was my earlier take on it. Hoogo's link has persuaded me it's real, but I agree whole-heartedly with you general sentiment, in particular that sticking a high end camera in Joe Blogg's hands does not necessarily a high end photographer make. Though he might do somewhat better, some of the time. The dpreview link also makes it clear the technique (at least for now) is quite limited in its scope: product shots (very very static) and maybe landscapes, but even then the slightest movement causes significant problems. Meanwhile, for scanners rather the opposite has happened, with all the major high end non-drum dedicated film scanner manufacturers bar one or two pretty much giving up the ghost, presumably because they think everyone has now scanned all their old film. They haven't! I would give a lot to have a technological advance that improved scanner dynamic range (and I suspect it's not a coincidence that the workaround, (true vertical) stacking of multiple different exposures of the same scan, is the workaround, but my own experience is that is more a triumph of hope over reality), but it ain't going to happen...

Oh, and those phones may be amazing - but I don't have one!

[bkh]

Thanks Hoogo, that's a good link, and I now get how it works, and furthermore can't disagree with the evidence in the OM crops of the tower - the level of detail is truly amazing (always was an OM man m'self, good to see they are still pulling the tricks). But, at the risk of being nit-picking, I still think the 'magic' is happening in the horizontal plane, by using the half pixel offset in that plane, and that talk of stacking, which by it's nature happens in the vertical pane, confuses the issue.

Just keep in mind that you have to upscale the original images by a factor of 2 first. Then the half pixel offsets translate into a 1 pixel offsets of the PL layers which add further detail to the final image. Just try Hoogo's example file: first, turn off the top three layers, then turn them on one by one to see the effect.

Cheers

Burkhard.

[Hoogo]

...I still think the 'magic' is happening in the horizontal plane, by using the half pixel offset in that plane, and that talk of stacking, which by it's nature happens in the vertical pane, confuses the issue...

I didn't check photographic forums for a long time now, not sure in what context "stacking" is used today.

"Focus stacking" might not be the only usage of a stack, but seems the term "stacking software" is widely used for this kind of stacking.
"Exposure stacking" became known as HDR, though it's not exactly the same.
"Noise reduction" by stacking multiple shots was once common.
"Super resolution" like we're discussing here can be seen as "stacking".
Then there's the stack in PS that gives a choice of the combination method. "Median" was famous for a while to let walking people disappear.
Different color filters is something for scientific stuff.

Guess we can find even more sorts of stacking.
Did someone use focus stacking for decreasing DOF yet?
What about shots with different angles of a polarization filter?