Stacking 10 images in P'shop to raise rez

[bruce1951]

I think, as a general rule, there is a trend to rely on technology to do the work. The days of doing your 'learning' are no longer relevant. It's a sign of the times! Instant gratification!! But what the engineers can never replace or even replicate is imagination and creativity.

The masters are those who do their own thing and don't rely on the latest gimmicks.

regards

[greenmorpher]

Fascinating discussion! I have asked the guy for a little more info, incljding whether this was a standard E-M5 shot (as shown in the EXIF) of an E-M5 II which had sensor shift to lift the pixel count to 40MPX from 16MPX, and what kind of P'shop stack increases resolution.

That sensor shift is the same sort of thing as Hasselblad had offered in the H5D-200c MS. (Did you know a Chinese drone company now has a majority shareholding in Hasselblad? They want the imaging technology, obviously -- high quality drone imaging has become hugely important at the top end of the market.)

Cheers, Geoff

[cathodeRay]

It is a fascinating discussion, thanks Geoff for starting it.

I was putting together a reply yesterday and then forgot about it and deleted it(!) but the gist of it was that notwithstanding Hoogo's stack of meanings of the word stack (and my gentle general remonstrance that photographers and digital photo editors tend to be rather sloppy about terminology - take lum* (all the brightness words) and saturation for example), the gist of 'stack' is putting one thing on top of another ie a vertical arrangement but then I remembered you can also 'stack the dishes' in a dishwasher and so it seems to me the essential gist is that the individual components are arranged such that the long axis of the stack passes though what is (usually) the wider plane of whatever it is that is being stacked. That is the essential feature, and the essential consequence is that they are in effect 'wired in series', so that whatever it is (light, electricity) that passes though them passes through them one after the other, in contrast to the parallel situation when whatever it is passes though all the items at the same time.

There is a clear similarity with the idea of two slides in the same projector vs the same two sides in two projectors pointing at the same point on the wall. Although in some ways they are similar, in others, notably effect, they are different.

Burkhard - I have downloaded and looked at Hoogo's example and yes I think I can see what is going on - but surely if the first step is to upscale, then we don't have any real new data, only interpolated data, some one or things best guess as to what might have been there? Getting back to the camera sensor shift: I can see how a sub unit level offset gets a different slant on things but it is still a horizontal offset, and the output increase is achieved in the horizontal plane (in effect an array of extra info is injected into the image) so that this is more of a merge (think of two traffic streams merging together, and of course generally when traffic streams are merging the only thing you absolutely want to avoid at all costs is stacking). Instead of three lanes of cars travelling in parallel, you now have five lanes travelling in parallel, and since we can imagine the cars in each lane are different colours, we also have more real data (five different points instead of three) at the instant we freeze the frame (make the picture).

Bruce - I absolutely agree about the false triumph of technology for technology's sake over artistry but I think I have to concede that what Olympus have done is pretty clever (like I said, I always was an OM man m'self back in the days, so it is good to see they are still pulling techno tricks). But then again, the technology, clever as it is, is somewhat limited in real world use, at least for now: product photography (hardly the real world), and at a pinch, very static landscape photography. The slightest hint of movement, and the technology goes phut and gives up the ghost (because it itself relies on movement, the offset). But at the end of the day the technology, useful as it can be, is more about 'wow' than 'ahhh... now I see'. To get the 'ahhh...' you need the artist behind the technology, and if the artist is good enough, the technology doesn't even have to be very high on the 'wow'. Plenty of superb photos have been taken on basic equipment...

The other potential risk with pursuing technology for technology's sake is you can end up chasing red herrings. Once you have a technology that can record (and recreate) all the meaningful data, then that's it. How many dpi do you need to get all the data off a 35mm slide, with any extra dpi adding only to file size, but not information? Can humans tell the difference between 16bit 44.1Khz and 24bit 96Khz recordings? This isn't about intermediate processing (when the maths may make more precision desirable), but the finished result.

One final thought: that OM zoomed in image of the lattice work on the tower on dpreview is definitely 'wow' (and it may also be pixel-peeping) - but how often do we use that detail that in our final image? How often do we print 8x10ft (ft is not a typo) images, and when we do, how often do we stand up close (where we can see the lattice detail) and how often do we stand back so we can see the wider picture - but not that lattice detail that we so laboriously gained?

cR

[bkh]

Well, if you don't like the "stack" metaphor, just pick a different one. Here "stack", imo, only refers to a "stack of images" to be processed together.

Burkhard - I have downloaded and looked at Hoogo's example and yes I think I can see what is going on - but surely if the first step is to upscale, then we don't have any real new data, only interpolated data, some one or things best guess as to what might have been there?

You don't have new data in the individual upscaled images, but each image contains different data, and what you do is merge data from different images. Upscaling is mainly necessary to fit the new data into the final image. You don't have to compute the final image the way Hoogo's example does, it's just an illustration of how it works in principle (using just an ordinary layer stack, nothing else).

Cheers

Burkhard.

[cathodeRay]

Burkhard - you are right, I was getting bogged down in being overly semantic. A stack to be processed is fine, like a stack of papers to be worked on.

cR

[Hoogo]

...OM zoomed in image of the lattice work on the tower on dpreview is definitely 'wow' (and it may also be pixel-peeping) - but how often do we use that detail that in our final image?...

"6MPixel are enough" was a slogan years ago, when manufacturers increased the resolution, shrinked the sensors and invented really a lot of noise. And that's quite true for the output picture. There's a nearest distance to look at a "whole" picture, get nearer and you don't se the whole thing anymore but details. No matter the size, it always ends with 6MPixel.

And there were the people that made prints of their negatives including the edges, to proof how well they thought about their picture. These should be perfectly happy with 6MPixel.

But isn't this way of looking at pictures "as a whole" not a bit restricted? It's fun to zoom into Gigapixel pictures, pictures of galaxies, stacked pictures of insects... And it's not really a new phenomenon that cam with the technology, that "fine art" stuff was quite a lot about catching most dynamic or resolution. There was hyper-realism in painting before.

HDR is widely accepted, and isn't this also a similar way of pushing the limits? Sunsets with silhouettes in the foreground are just fine, does it add some meaning when you open the shadows with HDR? What great pictures were made in the 30s and 40s with really low dynamic range.

And that 6MPixel rule did not really apply to the input side. You might collect stuff for montage, or zoom or macro aren't enough. The technology is just following the users, stitching and super resolution are done for years now.

[bruce1951]

OK another 'slant'.

We now have cameras that take multiple shots at different exposures that are used for HDR etc etc. Changing EV's. We have the subject of this thread. But as pointed out it's only applies to one very narrow focal plane. But for sometime now I have been waiting for manufactures to do 'focus' stacking. ie take multiple frames at different focus setting to produce a greater depth of field. Set up your camera as is done with exposure bracketing. But rather than changing exposure by set increments how about changing the focal length by set increments? I have used 'ControlMyNikon', tethered to a computer, to control the number of focus steps. So why can't this be done in camera? And if so wouldn't is get a better result than shift? More detail would be gained by changing the focal plane. Wouldn't it? And wouldn't those extra focal planes add to the pixel count and therefore the amount of detail? Imagine a DSLR on a tripod. Macro lens. Set up your focus steps and fire away. In camera stacking or use the captured frames to do your own stacking.

Hmmmmmm. Come on. I want a DSLR that will do focus bracketing.

bruce

[photoken]

...
And wouldn't those extra focal planes add to the pixel count and therefore the amount of detail?

Nope.

As I said, the blurred areas of each image are ignored (by sophisticated masking) when the images are combined into one "all in focus" image.

The pixel count remains about the same, but more detail is shown because of combining the details from the stacked images.

[photoken]

...
that OM zoomed in image of the lattice work on the tower on dpreview is definitely 'wow'

As soon as I realized that the "fuzzy" image was enlarged 158% to match the size of the towers, I lost all "wow" feelings....

[photoken]

...
Imagine a DSLR on a tripod. Macro lens. Set up your focus steps and fire away. In camera stacking or use the captured frames to do your own stacking.

That's already available on selected Panasonic cameras, including the compact ZS100. It's called "Post Focus", and here's the Panasonic blurb about it:

Burst images in 4K resolution (3840 x 2160 pixel) are shot at 30 fps while detecting max. 49 areas of a frame to set focus at different focus points – near to far. It is made possible by Panasonic high-speed, high-precision DFD (Depth From Defocus) auto focus technology.

Note that although it doesn't use the full 5.4K x 3.6K image size available when shooting single images, there's no reason that those burst images couldn't be combined in a focus-stacking app....

[Hoogo]

As soon as I realized that the "fuzzy" image was enlarged 158% to match the size of the towers, I lost all "wow" feelings....

Nice one, I didn't notice... Maybe scaling 200% and 300% would make results better comparable?

[bruce1951]

Ken I understand what you are saying about the deleted 'fuzzy' area of each frame. However you only have one fuzzy frame. That will stay relatively constant. Then you have all the 'extra' detail added as each frame is added to the stack. ie if I take a stack of a snake. The background is blurred/fuzzy in all frames. So that stays constant. But as each frame/plane of the stack, of which there could be hundreds, is added to the image, wouldn't that in affect be adding extra details therefor more data?
I'm not sure of how the Panasonic camera works. But my issue would be selecting the 'variable' focus plane/s which would depend on the subject. ie if I was photographing a stack of a snake I would dial in 30 frames/planes at 2cm intervals. (Depending how long the snake was!!!). However if I simply needed a stack for a landscape I may need just 4 frames/planes at a much great variation. What about a stack of a bee? Then I may want to dial in 30 frames at just .5 of a mm.
So you see I want a camera that will allow me to choose the number of frames and the distance/increment of each plane/focal shift. Make the selection/s and push the shutter button. Then sit back and watch as the camera fires of each shot and adjusts the lens! I don't want or need in camera stacking/processing. I want to have control of that.

regards

[photoken]

However you only have one fuzzy frame. That will stay relatively constant. Then you have all the 'extra' detail added as each frame is added to the stack. ie if I take a stack of a snake. The background is blurred/fuzzy in all frames. So that stays constant. But as each frame/plane of the stack, of which there could be hundreds, is added to the image, wouldn't that in affect be adding extra details therefor more data?

Yes, it adds more details, but No, it doesn't add more data.

For any given point in your image rectangle, the pixel at that point has data -- it might be blurry data or it might be "sharp" data, but each point in your image rectangle always has some kind of data. Focus stacking is not adding more data, it's just changing the data of a pixel from blurred to sharp. In other words, if you start with a stack of 20Mpx images, you'll end up with an image of 20Mpx.

I'm not sure of how the Panasonic camera works. But my issue would be selecting the 'variable' focus plane/s which would depend on the subject. ie if I was photographing a stack of a snake I would dial in 30 frames/planes at 2cm intervals. (Depending how long the snake was!!!). However if I simply needed a stack for a landscape I may need just 4 frames/planes at a much great variation. What about a stack of a bee? Then I may want to dial in 30 frames at just .5 of a mm.
So you see I want a camera that will allow me to choose the number of frames and the distance/increment of each plane/focal shift. Make the selection/s and push the shutter button. Then sit back and watch as the camera fires of each shot and adjusts the lens! I don't want or need in camera stacking/processing. I want to have control of that.

Yes, you're correct about different subjects needing different amounts of frames. In fact, I've seen sample images from the various focus stacking sites that follow exactly what you describe -- 3 images for a landscape, and 20 or more images for a shotgun photographed end on, and probably even more images for a head shot of an insect.

I don't see why your idea of those choices couldn't be implemented, although it would be hugely complicated and time-consuming to set up, I'm thinking....

The way that Panasonic does its "Post Focus" thing is that it divides the image frame into 7 rows of 7 columns and does its burst of images with each of the 49 areas getting the focus. Very simple and dead easy to do -- point, click, and two seconds later you've got your images. Whatever the type of subject, there's no muss and no fuss.

[bruce1951]

OK Ken. I see. A 6000x4000 image will always have 6000x4000 pixels. Good thinking. Blurry or not.

As for focus bracketing all I want is control. Software such as ControlMyNikon is idea. Just get rid of the darn cable and computer!!

regards

[Hoogo]

Not sure when it can be calculated like this.
If you take 4 equal pictures to reduce noise, you have reduced noise, so there's more data.
HDR is similar. Same resolution, the extra data is hidden in the dynamic.