Superresolution - A Photoshop technique for enhancing image resolution.

[John 2]

Found this today and I will certainly try it because it looks quite effective. Preferably needs PS CS6 and above but a work around for other versions is described. I would guess that it also requires a lot of memory. We shall see. It's based on combining multiple shots and so it is not that suitable for scenes containing movement.

Thought it worth sharing.

http://photoncollective.com/enhance-...dobe-photoshop

[FrankMi]

Really neat article, John.

I need to give that a try!

[John 2]

I agree Frank. There seem to be two possible benefits, resolution and noise. Certainly worth a try for the right image. If you do give it a try, post the result.

[ajohnw]

It would probably be easier to do with astro software - all he is really doing is averaging various types of noise via a lot of shots.

Super Resolution - bull stuff even from hasselblad. It probably does allow a larger 100% resolution image / improves interpolation up in size but increase the actual resolution of the lens which is the real limiting factor - don't think so. It might help debayering if done on raw files before they are processed at all. I undestand Pentax did this for another reason on one camera - no need for an anti aliasing filter so no moire.

John
-

[Manfred M]

bull stuff even from hasselblad.

Agree with you 100% John.

All they are doing is effectively increasing the apparent size of a sensel of data , "losslessly" upscaling the resultant product and then downscaling back to 100%; nicely said, sharpening the image. I'm willing to bet I can accomplish the same effect in post without ever stepping outside with a camera. Maybe I'll try this and post the results.

Another example of marketing BS (by Hasselblad and if the article is correct, Olympus) or in the case of the article, another photographer who does not understand what is happening behind the scenes coming to the conclusion you can break the laws of physics. Maybe we can sell this guy a perpetual motion machine.

[ajohnw]

Some astro people do dither shots Manfred and higher resolution has been claimed for that too - there have been cases of shots where it has been said to have been used where the telescope couldn't possibly resolve what it has. I have some microscope software that removes spherical aberration or says it does.

I can see how it might work but moving the camera in pixel increments would blur or do nothing if they were aligned.

I wouldn't knock Olympus too much - they are cavalier enough to change their raw processing software - on the other hand in real terms they are probably just following suite. No doubt some where on the web the results will be great. They are achieving rather high resolution figures on some of their lenses but I would still have to wonder due to no anti aliasing filter and moire.

John
-

[Manfred M]

John - I'm not knocking anyone other than the marketing departments of camera makers that promise something for nothing; I think virtually all of them are guilty here. After all, if Hasselblad gets away with it, why not others?

As for the dithered astro shots; people who don't understand what is happening with the routines can (and do) feel that when something looks better, it must be more accurate. It's no different than when the mFT manufacturers (both Olympus and Panasonic are guilty of this) embed lens correction data in the RAW data and represent what is essentially fiddled with data as what the camera has recorded. Ditto with CA removal and lens vignette removal; both effects are there and it only goes away through fiddling with the pixels.

I always laugh when some SOOC photographers get on their high horse and tell us that those of us who PP our images are cheating, while they are the purists. Little do they understand how much fiddling has gone into their jpegs.

Just a historical note: Have a look at this early 17th century El Greco painting:

http://www.elgreco.net/christ-carryi...sp#prettyPhoto

Look at where Christ's hands meet the cross and the robe. I see a halo; it looks like "sharpening" techniques have been with us since the 1600s.

[James G]

This is not really anything new. It's basically noise removal using opacity stacking.

I used to use it when stacking shots of the moon. In this instance, I used to stack 20 or so shots.

For obvious reasons the camera and telephoto was tripod mounted.

The technique depends on the noise not being common to each frame. Fine in principle, but actual experience was that the final stack was invariably softer than any of the original frames. I came to the conclusion this was because atmospheric distortion actually changed the absolute dimensions of the moon in each frame by a few (sometimes more than a few), pixels.
As a result fine detail, visible despite noise in the individual frames, disappeared. (presumably because it starts to resemble real noise.)

At the end of the day it was better to process individual frames for noise and properly apply selective sharpening using luminosity masks to target the fine detail.

[chauncey]

FWIW, I will often shoot multiple shots of the same scene using AF on each shot, then use normal
Photostacking in PS CC with, oftentimes, impressive results.

[PhotomanJohn]

I have no experience with this but here is a link to PhotoAcute's web page:

http://www.photoacute.com

I have the book that is mentioned at the bottom of the webpage (Photographic Multishot Techniques by Juergen and Rainer Gulbins, published by rockynook) which has a detailed chapter devoted to using this software and multiple exposures to produce "super resolution" photos. I read this book some years ago and the results seemed to be legitimate.

John

[Shadowman]

Nice link.

[Richard Lundberg]

For any who doubt, super resolution is an outgrowth of the generalized sampling theory of Athanasios Papoulis, author of my grad school text "Probablity, Random Variables, And Stochastic Processes. There are some quibbles about existence, stability, etc, but there are many subsequent papers and applications. Google will quickly fill your desktop with pdfs.

[John 2]

Slightly surprised at some of the reaction to this post. As a pure consumer, I am not qualified to comment on where manufacturers have tried and failed but the main thrust to me seems to be a none profit making tutorial which on the face of it, demonstrates some benefit. As a consumer, there can only be one test - suck it and see.

[ajohnw]

For any who doubt, super resolution is an outgrowth of the generalized sampling theory of Athanasios Papoulis, author of my grad school text "Probablity, Random Variables, And Stochastic Processes. There are some quibbles about existence, stability, etc, but there are many subsequent papers and applications. Google will quickly fill your desktop with pdfs.

There is a decent run down here. It shows how what is effectively averaging can improve things but that hasn't increased resolution just made aspects clearer. There is also a paragraph on under sampling effects. The paragraph on sub pixel "movements" is a bit glib.

http://en.wikipedia.org/wiki/Superresolution

Where lots of work on super resolution with real outcomes that get past the diffraction limit have been successful in real terms is partly run through here. It's interesting to note the "single item" mentioned on one of them.

http://en.wikipedia.org/wiki/Super-r...ion_microscopy

The evanescent wave one might be of interest to macro photographers if they can get the lens close enough. I've often wished I could buy a few glass beads of the right size as it's easily done on a normal microscope. The bead re images the subject so that it can be examined via the microscope. It seems this effect was known very very early on. I've always wondered how.

In terms of photography I suppose the circle of confusion has similarities to the diffractions spot but will be far less controlled in real lenses but pixels are hidden behind debayering which is an averaging/interpolating thing. That also probably contains modifications that dsp people call cad - things that work that were obtained by trial and error rather than any classical theory. I can see how it could help that. I'm also inclined to think of slant edge lens resolution tests which make use of sub pixel measurements - the resolution tends to improve when tested on a camera with more pixels. Best leave people to think about that and wonder why the image might also appear sharper which isn't the same thing as saying resolution has increased.

John
-

[Manfred M]

Slightly surprised at some of the reaction to this post. As a pure consumer, I am not qualified to comment on where manufacturers have tried and failed but the main thrust to me seems to be a none profit making tutorial which on the face of it, demonstrates some benefit. As a consumer, there can only be one test - suck it and see.

John - there are a few of us who have the technical background to understand what is going on with some of these "features" and fairly quickly figure out what makes sense technically, and what is "smoke and mirrors".

I spent much of my career working with the marketing groups supporting new product development and roll out. One of the most "feared" meetings was when they wanted to have more "new" or "new and improved" features to market the products with. Fortunately, the companies I worked for had legal departments that forced the marketing folks to stay within the letter of the law; i.e. the claims could be substantiated in front of a third party, if need be (i.e. the courts).

That being said, the law does not hold that these "features" have to be useful to the end user; and this is very much where this turning a 50MP sensor into a 200MP with a bit of pixel shuffling (and in my view it truly is a stretching the truth to the breaking point).

I've owned the Guilbins book for a few years, but had not looked at that particular chapter. A few facts; Photoacute came out in 2004 and has been updated since. Gulbins was using a 10.1MP Canon 40D when the book was written, it came out in 2008. So while the technology might have been relevent then; in reality, there is a far more limited need, even for the pixel peepers. The technology should work well in reducing noise, but any other gains are rather useless as they are nothing more than smoke and mirrors (effectively a form of sharpening).

[John 2]

Don't get me wrong Manfred. I wouldn't question anybody's knowledge on the subject. I am (was) a trained engineer but spent most of my working life writing legal contracts and so I understand fully the points that you make. However the examples in his video seem to show that the process is capable of benefitting the right sort of image however that is achieved (I choose my words carefully). That being so and since the guy isn't asking me for money, I don't want to dismiss it until I have tried it for myself and found it wanting.

[Manfred M]

Don't get me wrong Manfred. I wouldn't question anybody's knowledge on the subject. I am (was) a trained engineer but spent most of my working life writing legal contracts and so I understand fully the points that you make. However the examples in his video seem to show that the process is capable of benefitting the right sort of image (I choose my words carefully). That being so and since the guy isn't asking me for money, I don't want to dismiss it until I have tried it for myself and found it wanting.

John - I did go out and try a variant of the approach; using a technique that I suspect is similar to what Hasselblad and Olympus seem to be doing with a pixel shift approach.

The website that suggests multiple handheld exposures is simply too unscientific and unrepeatable to be useful for analysis purposes. I went out and got a reasonably sharp image that I shot on the D800 and did a one pixel shift in each direction. This means I ended up with 9 separate, but slightly different images, with a 2 pixel shift in the x and y directions, much as I would expect to get if I shot on a tripod with either the Hassy or the Oly cameras.

I assembled the image exactly the same way as the website recommended, using the same parameters that the author used. I did not publish the pictures because at the end of the process, there was NO discernible difference between the original image and the super-resolution image. My test for this is to take the two images, stack them one on top of the other and then use the difference or subtraction blending mode to reveal any differences between the two layers.

End result - there is no difference in my test in IQ between the original and the one that had the work done to it. That is a bit of a surprise, but then, perhaps not; perhaps things would be different with a lower res camera?

[James G]

Manfred had kind of beaten me to it.
I too tried the technique, and could see no advantage to it.

Attached are 4 images for perusal. ( If anyone wants, I will put the full size originals and 'processed' results up on my Skydrive and post a link).....

#1 Source image from ACR (Photoshop) (one of 10 handheld shots that were subsequently stacked and blended)



#2 Source image from #1 with additional Smart Sharpening



#3 result of all 10 images, processed using suggested technique with same level of Smart Sharpening as for #2



#4 result of 'standard PS stacking and merging with same Smart Sharpening as for #2



To all intents and purposes the result from the standard process in PS is indistinguishable from the 'Super Resolution' process.

More significantly perhaps is the fact that just processing a single exposure correctly gives more or less the same result (I have not tried pixel peeping to gauge noise levels)

The images were captured in our conservatory, using a Canon EOS 7D Mk2 with a Sigma 17-70mm lens attached. No flash and relatively 'dull' conditions.

Part 2 will follow in another post and is rather surprising .....

[James G]

OK... part 2...

I mentioned on an earlier post in this thread that I had tried using Stacked Opacity blending on moon shots and not been particularly impressed.

actual experience was that the final stack was invariably softer than any of the original frames. I came to the conclusion this was because atmospheric distortion actually changed the absolute dimensions of the moon in each frame by a few (sometimes more than a few), pixels.
As a result fine detail, visible despite noise in the individual frames, disappeared. (presumably because it starts to resemble real noise.)

I was intrigued by the idea of increasing the image sizes by 200% , and using 'nearest' neighbour sampling' before aligning and subsequently applying the opacity blend changes. (It had never occurred to me to do this. Additionally, I had always applied Vignette and Geometric Distortion checks)

So I hunted out some recent moon shoots and applied the Super Resolution Technique out of curiosity....

Results as follows....

#1 This is the result of stacking 10 captures but NOT initially increasing image size by 200%.



#2 This is the result of stacking the same 10 captures, (from #1) and increasing image size by 200%.




and just in case this was a flook... I ran a second series...

#3 Again this is the result of stacking 10 captures but NOT initially increasing image size by 200%.



#4 And this is the result of stacking the same 10 captures, (from #3) and increasing image size by 200%.



In both cases I seem to have an image with significantly more clarity/definition, using the upsizing process.

Both series were sharpened using Smart Sharpen.I do not normally use Smart Sharpen much, preferring to use my one sharpening recipes, so I substituted my own 'preferred' sharpening and generated similar results.
(I normally use a combination of Gaussian, Surface Blur and noise/median frequency actions to apply specific sharpening to masked areas of an image.)

Regardless of the final sharpening method, the end result was the same.

At the moment I'm puzzling why the effect it is so much more evident with this type of shot compared to the 'normal' scene I used to test initially.

???

(Stating the 'obvious' none of these images where converted to grayscale )

[Shadowman]

OK... part 2...

I mentioned on an earlier post in this thread that I had tried using Stacked Opacity blending on moon shots and not been particularly impressed.


I was intrigued by the idea of increasing the image sizes by 200% , and using 'nearest' neighbour sampling' before aligning and subsequently applying the opacity blend changes. (It had never occurred to me to do this. Additionally, I had always applied Vignette and Geometric Distortion checks)

So I hunted out some recent moon shoots and applied the Super Resolution Technique out of curiosity....

Results as follows....

#1 This is the result of stacking 10 captures but NOT initially increasing image size by 200%.



#2 This is the result of stacking the same 10 captures, (from #1) and increasing image size by 200%.




and just in case this was a flook... I ran a second series...

#3 Again this is the result of stacking 10 captures but NOT initially increasing image size by 200%.



#4 And this is the result of stacking the same 10 captures, (from #3) and increasing image size by 200%.



In both cases I seem to have an image with significantly more clarity/definition, using the upsizing process.

Both series were sharpened using Smart Sharpen.I do not normally use Smart Sharpen much, preferring to use my one sharpening recipes, so I substituted my own 'preferred' sharpening and generated similar results.
(I normally use a combination of Gaussian, Surface Blur and noise/median frequency actions to apply specific sharpening to masked areas of an image.)

Regardless of the final sharpening method, the end result was the same.

At the moment I'm puzzling why the effect it is so much more evident with this type of shot compared to the 'normal' scene I used to test initially.

???

(Stating the 'obvious' none of these images where converted to grayscale )

Nice comparisons. I think the reason for the difference is that the first series didn't need as much sharpening, there was nothing to correct.