Previous news story    Next news story

Adobe admits using 'synthetic blur' image in deblur demo

By dpreview staff on Oct 18, 2011 at 22:00 GMT

Adobe has admitted an image used in its 'image deblur' presentation was artificially blurred for the purposes of the demonstration. The company said the blur on the image was 'more complicated than anything we can simulate using Photoshop's blur capabilities.' It described the move as 'common practice in research' and defended the use of the image because 'we wanted it to be entertaining and relevant to the audience.' The other images shown were the result of camera shake, it said.

The original, unshaken, image of Adobe's Kevin Lynch by Kendall Whitehouse

The admission came as part of an article on the company's Photoshop Blog, that describes the technology in more detail.

Adobe's statement on the issue, from the Photoshop Blog:

'The first two images we showed – the crowd scene and the image of the poster, were examples of motion blur from camera shake. The image of Kevin Lynch was synthetically blurred from a sharp image taken from the web. What do we mean by synthetic blur? A synthetic blur was created by extracting the camera shake information from another real blurry image and applying it to the Kevin Lynch image to create a realistic simulation.'

'This kind of blur is created with our research tool. Because the camera shake data is real, it is much more complicated than anything we can simulate using Photoshop’s blur capabilities. When this new image was loaded as a JPEG into the deblur plug-in, the software has no idea it was synthetically generated. This is common practice in research and we used the Kevin example because we wanted it to be entertaining and relevant to the audience – Kevin being the star of the Adobe MAX conference!'

Comments

Total comments: 97
12
crustyjuggler66
By crustyjuggler66 (Oct 24, 2011)

They must have done something similar for 'Content Aware' cause that's never behaved liked the videos showed it would.

Sad tactics..

0 upvotes
Detail Man
By Detail Man (Oct 23, 2011)

See this related post on the DPR News Discussion Forum discussing Adobe's latest move:

http://forums.dpreview.com/forums/read.asp?forum=1000&message=39659342

0 upvotes
ProDesignTools
By ProDesignTools (Oct 23, 2011)

Of interest: Some brand new before-and-after images from the tool have just been released, along with further details

http://prodesigntools.com/photoshop-image-deblur-adobe-releases-new-photos.html

The historical B&W Capa D-Day photo is a notable inclusion. Independent submissions are also invited

Comment edited 3 times, last edit 2 minutes after posting
3 upvotes
GBo
By GBo (Oct 24, 2011)

Thanks for the link

2 upvotes
Detail Man
By Detail Man (Oct 23, 2011)

Photog23 wrote:

"But this fraud was great marketing. When I go out to the web and look for the "correction," to their fraudulent presentation, I get very few links." ...

I guess that "money talks", thus enabling "bullcrap to walk" ?

... "But I get raftloads of links to pages about the wonders of Adobe's new deblur technology. The Adobe marketeers work is done."

Indeed. If you can only manage to hook "some of the people all of the time", that is (monetarily) more than enough to live off the "fat of the land" (so to speak). The triumph of "form" over actual function ...

Annual re-licensing of Photoshop will constitute the final enslavement of the soul to "The Monolith". "The first time is free" - but the withdrawal symptoms can be deadly. You *know* that you need it to be real ...

Rogue gangs of moribund and toothless zombies will soon stalk the earth in legions, "fencing" municipal infra-structure common metals to smelting plants - all just for that next Adobe "fix" ... ;-)

0 upvotes
photohounds
By photohounds (Oct 25, 2011)

A bit like the Japanese marketing assault on Chromium Dioxide tape in the 1970's.

Patented by BASF, (i think) the Japanese electronics giants didn't want to pay royalties ...

BS - but it planted the seeds of excessive head wear when the REVERSE was actually true.

I read in some IBM publication in the late 1980s, that they used Chrome tape in their mainframe backup drives - BECAUSE of low head wear.

0 upvotes
photohounds
By photohounds (Oct 25, 2011)

People who don't research will be disappointed - too bad for them.

Still ... they cam always start a class action if they choose to.

0 upvotes
Photog23
By Photog23 (Oct 21, 2011)

I consider this a relief! This deblurring technology would have been a compelling reason to upgrade to the next version of Photoshop if it was offered there! Now, the only way I'll be convinced that this technology is real is if they take a batch of my blurred images and convert them to my satisfaction.

But this fraud was great marketing. When I go out to the web and look for the "correction," to their fraudulent presentation, I get very few links. But I get raftloads of links to pages about the wonders of Adobe's new deblur technology. The Adobe marketeers work is done.

Many of Adobe's technologies are smoke and mirrors. For instance, the refine edge tool seems amazing, until you consider that every single person demonstrating it extracts their images from stark monotone background. Take a picture where you have a real reason to use the tool? Doesn't work very well. You got to be very careful with Adobe and always peek behind the curtain ;-).

1 upvote
Detail Man
By Detail Man (Oct 20, 2011)

Adobe's intentional blurring of the truth revealed:

http://blogs.adobe.com/photoshopdotcom/2011/10/behind-all-the-buzz-deblur-sneak-peek.html#comment-12880

Comment edited 34 seconds after posting
1 upvote
Detail Man
By Detail Man (Oct 20, 2011)

BREAKING AND TRULY SHOCKING NEWS:

http://forums.dpreview.com/forums/read.asp?forum=1000&message=39638642

1 upvote
lbjack
By lbjack (Oct 20, 2011)

"A synthetic blur...to create a realistic simulation."

Yeah, right.

0 upvotes
Fine Art
By Fine Art (Oct 19, 2011)

The mission to fix the Hubble telescope cost 1.5 Billion $. If all the braniacs at NASA couldn't do it in software for a billion what are the chances it will be in your photoshop upgrade for $300?

Presenting a perfect deconvolution as something you will get in your consumer software is plain fraud. Deconvolution is real, it works. I use it a lot. It is not going to do miracles.

You can buy deconvolution in other software now. I recommend Images Plus. Ive been buying it since version 2. I get nothing for recommending it.

0 upvotes
Victor Engel
By Victor Engel (Oct 19, 2011)

The problem with the Hubble wasn't motion blur. It was focus blur. There's a big difference between the two. In the case of motion blur, the picture is actually in focus (well, it doesn't need to be, but then there are two kinds of blur).

0 upvotes
Fine Art
By Fine Art (Oct 19, 2011)

Have you ever actually done a deconvolution?

0 upvotes
DSPographer
By DSPographer (Oct 20, 2011)

Actually, the Hubble problems spurred the refinement of one of the most popular deconvolution methods "Richardson-Lucy". The limitations of deconvolution meant that it was still worth the huge expense to fix the telescope. Note: you probably used a variation of this algorithm since the space telescope science institute made the code available without restrictions. Here is one article about it:
http://adsabs.harvard.edu/full/1994ASPC...61..296S
The current Wikipedia article has a nice brief introduction to deconvolution including blind deconvolution:
http://en.wikipedia.org/wiki/Deconvolution

1 upvote
Fine Art
By Fine Art (Oct 20, 2011)

I do use an adaptive RL.

0 upvotes
ccr294
By ccr294 (Oct 19, 2011)

If Adobe had the integrity that I thought they had (yes I'm naive) they would have done what a previous poster indicated - took one clear image & one blurred image of the same subject. Then posted the images with indisputable documentation.

I do feel deceived and it will be a long time before I forget.

If this process really works, Adobe has a chance to redeem themselves if they feel it warranted and take a "do-over" (pun intended)

And as the poster commenting on the 'slow computer' - really! that was some weak-assed crap.

1 upvote
Joseph S Wisniewski
By Joseph S Wisniewski (Oct 19, 2011)

You said "it will be a long time before I forget."

I say "good!"

People shouldn't forget.

This is the Adobe that had programmer Dmitry Sklyarov jailed for making an eBook reader for the blind, something that fell well within any accepted definition of "fair use".

This is the Adobe that needed some positive press after dropping their open PhotoShop plugin SDK and going to a new system where you paid for the SDK, but they might keep your money and not send you an SDK, so they slapped together the apparently useless DNG "open" standard and started dancing around saying "we're not the bad guys getting people arrested for decoding files: we're the champions of open standards!"

This is the Adobe that announced a lightfield camera of their own, to compete with the nebulosity of Lytrol.

This is the adobe that wants to move to a "yearly rental" licensing scheme for PhotoShop.

3 upvotes
Joseph S Wisniewski
By Joseph S Wisniewski (Oct 19, 2011)

As far as the "slow computer", I'm guessing that's any desktop or laptop computer. Fast, in terms of algorithms like this involves a small rack of blades at the deskside, or several full sized racks in the next room.

0 upvotes
photohounds
By photohounds (Oct 25, 2011)

Indeed, Joseph.

People forget only too quickly - the "next big thing" covers all past transgressions it seems.

I use Bibble (purchased) for workflow and GIMP for the more detailed work.

Bonus: they run on multiple platforms so I'm not tied ....

1 upvote
edfo4
By edfo4 (Oct 19, 2011)

If it pans out, it's a really great feature. Too bad that Adobe had to resort to trickery in its presentation. What rankles the most however is the presentor's comments about a "slow" computer. Really, guys, come on!

0 upvotes
MGJA
By MGJA (Oct 19, 2011)

Sorry, Adobe. That the picture you were caught with red-handed is the only one that got... uh, digitally enhanced for greater marketing impact, while the other two are real, REAL, cross-our-marketeering hearts is kinda hard to swallow.

Unless you post the original RAW files - preferably with a pre-demo time stamp from a service you don't own directly, like Flicker - your credibility on this issue is zero.

Maybe you have a winner on your hands here. Who knows? But until your efforts have been reviewed and verified by independent third parties, the rational thing to do at the moment, given your conduct, is to assume this is all so much exaggerated hype for the sales drones to regurgitate.

1 upvote
Joseph S Wisniewski
By Joseph S Wisniewski (Oct 19, 2011)

"the other two are real, REAL, cross-our-marketeering hearts is kinda hard to swallow."

Actually, it's not. If you run hundreds of images through an algorithm like that one (who cares if it's slow, Adobe has a surplus of interns) you can "cherry pick" a couple that have just the right blur to look good when processed. So what if, in real life, 99.7% of images don't look "photographic" after deconvolution...

1 upvote
Kerensky97
By Kerensky97 (Oct 19, 2011)

"The first two images we showed – the crowd scene and the image of the poster, were examples of motion blur from camera shake."
Am I the only one reading this that the first two pictures were legitimate (or the only one reading more than the headline of this article?)

If you watched that and said "WOW!" for those first 2 images you're still talking about a real life application. The last one may have been a setup but from what I'm seeing this is still a 100% viable feature. Even if it's not as impressive as the manipulated shot of Kevin it still made huge improvements on the other shots.

Maybe people should see this application in use first hand before they jump to conclusions that "Adobe lies! Photoshop is just MSPaint marked up to $300!"

2 upvotes
duartix
By duartix (Oct 20, 2011)

What does the saying go like?
"Lie to me once...shame on you, lie to me twice, shame on me..."?

So they plain lied to everyone about the last image, the first image has notably strange blur characteristics (like I've never seen before on a human shaken photo) and you still think it is legit???

Well I'm sorry to say, but shame on you...

1 upvote
wetsleet
By wetsleet (Oct 20, 2011)

@Duartix
Spot on. The more I look at the supposedly 'real' camera shake pictures, the less they look like any of my own camera shake pictures (I have many :< )

No, they look much more like two sharp pictures overlaid slightly out of register. Just look at the Oakley 'O', or the 'GUESS' in the crowd scene, or any of the text in the holiday poster. It's like they had the camera on a robotised platform programmed to move the camera during the exposure with about 90% of the exposure static at either end of the trajectory, and a very rapid transition between the two ends.

OK, so it is a controlled starting point. But why do they dissemble, and give the impression that here are just two everyday handheld photos which went wrong, and we fixed them?

Given the suspension of disbelief required to accept their presentation, it must have been obvious that when the pazzaz died down people would start to question what they had seen. Much better to have laid out all the caveats honestly in advance

1 upvote
Robgo2
By Robgo2 (Oct 19, 2011)

What?! A giant corporation misrepresented the facts for financial gain? Shocking! Absolutely shocking!

Actually, this is not unlike the anti-ballistic missile test run by the U.S. military, in which the target missile was emitting a tracking signal. Without that signal, the ABM was unable to hit the target. Perhaps Adobe will eventually get their technology to work, but they should not claim success until they can demonstrate its effectiveness under realistic testing conditions.

Rob

3 upvotes
duartix
By duartix (Oct 19, 2011)

I can't think of any other word for this other than "Cheating", let me tell you why...

Estimating the PSF/debluring kernel is only one part of the job and so far it looks like they have accomplished it (not that I trust this any more), however inherent image noise is one of the biggest barriers for deconvolution filters and it usually results in all kind of wild sharpening artefacts.

Using a pre-blurred image is obvious cheating because they removed the noise problem out of the equation.

6 upvotes
roy5051
By roy5051 (Oct 19, 2011)

I would imagine that it would be quite easy to remove blur you had added yourself, wouldn't it?

0 upvotes
ZorSy
By ZorSy (Oct 19, 2011)

Thanks, I knew I'd seen this somewhere before - it's called "UNDO" on my computer.....

8 upvotes
BitFarmer
By BitFarmer (Oct 19, 2011)

The only way to automatically score different debluring algorithms is to apply them to artifially blurred shots so you can make a direct comparation between the original sharp one and the one sharped by the algorithm.

Using real blured shoots may seem more fair, but inusable for scoring just because you have nothing to compare with.

2 upvotes
duartix
By duartix (Oct 19, 2011)

@BitFarmer:

You may be well intentioned but what yo are saying it's not true. In a controlled environment you can easily take two comparable pictures (one blurred and one sharp) by either using different ISO/shuttter speed combinations or by using a tripod or a human.

They are using one(?) artificial blurred picture because they are probably concentrating on one problem at a time and deferring the noise barrier for later nuptials.

1 upvote
epo001
By epo001 (Oct 19, 2011)

The picture wasn't what people thought Adobe said (or alleged) it was, this will make people mistrust future claims they make for this (c.f. Topaz Labs InFocus)

What they did was understandable, by extracting the blur info from a known (and presumably successfully processed) image they were playing it safe by choosing something which the software already knew how to process.

All demos are rigged, you would be insane to risk a belly flop in front of a live audience but you have to be careful not to mislead the audience. This technology may or may not have promise but Adobe now have to overcome the hurdle of these seemingly exaggerated claims.

Comment edited 2 times, last edit 4 minutes after posting
2 upvotes
CAClark
By CAClark (Oct 19, 2011)

My only gripe is the guy stood there and said it was a crappy bad photo they got on the internet. All they had to do was not embellish the facts.

9 upvotes
WT21
By WT21 (Oct 19, 2011)

We'll have to see if it really works, but that's just terrible not to disclose the nature of the picture in the demonstration.

Comment edited 50 seconds after posting
0 upvotes
ZorSy
By ZorSy (Oct 19, 2011)

http://forums.dpreview.com/forums/read.asp?forum=1034&message=39616976

enough said

0 upvotes
a1man
By a1man (Oct 19, 2011)

Great catch

0 upvotes
IanGilbert
By IanGilbert (Oct 19, 2011)

As a1man says, it is a great catch. The examples posted online and in print ads by software makers are entitled to zero credibility. The software does what independent reviewers say it will do, not what the manufacturer claims ("Dreck gets your laundry three times cleaner in half the time!" Right.).

The same is true for manufacturers' images demonstrating their newest cameras and lenses -- all you need to get the same results are three assistants, 500 pounds of lights, reflectors, and shades, and a 200-pound tripod. Then you select one image out of 3,000 and add a banner headline, "Look, Ma, no hands!"

Comment edited 2 minutes after posting
1 upvote
maiaibing
By maiaibing (Oct 19, 2011)

Sound disbelief can be a very good thing. Already. I have this strong gut feeling that I will not upgrade to the next edition of PS just to get their new magic anti-blur technology...

0 upvotes
Charrick
By Charrick (Oct 19, 2011)

First of all, I think that this technology has great potential. But it's sad that Adobe somewhat ruined it by using this synthetically blurred photo. That only breeds suspicion. Perhaps the first two photos were truly average blurred photos, but it's harder to take them completely on their word. I hope it's true, though.

One other thing. Did anyone notice the Adobe guy apologizing for the "slow computer". College kids showing each other cool programs might do that, but one of the most technologically oriented companies in the world can probably afford a computer that's at least an above average consumer PC when they're showing off their new software. It seems obviously to me that they got the best computer that money can buy and then apologized for it being slow.

Also, those idiots in the chairs on stage were completely annoying.

With all of this said, if that technology is actually as was demonstrated, it is one of the greatest advances in photo editing in history.

2 upvotes
Anssi Kumpula
By Anssi Kumpula (Oct 19, 2011)

Exactly. If that technology delivers, great. But fooling people and unprofessional presenting is bad.

1 upvote
Anssi Kumpula
By Anssi Kumpula (Oct 19, 2011)

What Adobe did is presenting a non-real world photo that was not real-world photo. Was that a mistake?

I think it was, even if it doesn't make that technology look very promising.

0 upvotes
Anssi Kumpula
By Anssi Kumpula (Oct 19, 2011)

Seems like I can't write. What I meant is Adobe using a non-real world photo in a demo but presenting it as a real world photo and that was a mistake.

The guy in the demo said: "Okay one more comparison. We found this image online. It's random internet photo... it's bad." Which clearly is to claim that photo was genuinely blurred by out-of-focus, camera shake or subject movement.

1 upvote
G Davidson
By G Davidson (Oct 19, 2011)

Please people, have a sense of humour. The people making photoshop, Lightroom etc are just that- people. Forget about bashing the 'big monolithic company' for 5 minutes and appreciate what they have made for us.

2 upvotes
JWest
By JWest (Oct 19, 2011)

They're just people? That changes things then. I already expect huge profit-hungry corporations to constantly lie to me - that goes without saying. But individual people who lie to me? They're the lowest of the low.

1 upvote
bobbarber
By bobbarber (Oct 19, 2011)

For us? Where's my dividend check?

1 upvote
luap_42
By luap_42 (Nov 18, 2011)

Bullcrap G Davidson

0 upvotes
Denis James Evans
By Denis James Evans (Oct 19, 2011)

honesty is the best policy. When you loose credibility it takes a long time to get it back.

2 upvotes
Fred Dominic
By Fred Dominic (Oct 19, 2011)

I don't know, the synthetic image doesn't bother me that much, it is research and the product is not ready yet, and it is not as if they are offering it for sale yet. Often people will test their algorithms on synthetic data first, before testing it on real data. I'm certain that they won't include it in CS6 if it doesn't work on real images.

Is no one bothered, however, by the ridiculous person who was on stage at the same time of the presentation, the person going on about what an "algorithm" is, and that "you should sell this"? I found that distracting, insipid, and technically profane.

2 upvotes
huyzer
By huyzer (Oct 19, 2011)

He wasn't too funny, I agree. But being caught like Adobe was, isn't too funny either.

0 upvotes
wetsleet
By wetsleet (Oct 19, 2011)

"Often people will test their algorithms on synthetic data first"

But that is not the whole story here. It would be one thing to work on the synthetically blurred image from cold, quite another to have access to and make use of the parameters which governed the creation of the blur in the first place.

Divining those parameters from the blurred image alone (synthetic or otherwise) is half the battle, at least.

I have seen suggestions that those parameters could eventually be sourced from in-camera accelerometers, added to the EXIF at the time, obviating the need later to winkle them out of the image from cold. That would legitimise what has happened here.

But then you must ask, aren't those the same accelerometers and the same data which drive mechanical VR anyway. To then suppress the mechanical VR only to replace it with a lesser 'software VR' makes limited sense.

In all events, leaving out that significant detail from the presentation is a major faux-pas.

Comment edited 2 times, last edit 3 minutes after posting
0 upvotes
arboreus
By arboreus (Oct 19, 2011)

Algorithms of deblurring are known for a long time already:

http://www.cse.cuhk.edu.hk/~leojia/projects/motion_deblurring/index.html

And yes, in research images are often artificially blurred with a known kernel ("camera shake data") to be able to control also the intermediate results of the algorithm.

The real question is how much CPU time and memory one needs to deblur a 15MP image.

Comment edited 37 seconds after posting
2 upvotes
gl2k
By gl2k (Oct 19, 2011)

In 2 years everyone will have : 6 cores hyperthreaded CPU with at least 12 Gig of RAM. Smartphones will have quadcore CPUs with 2 Gig RAM.
I bet the latest smartphones could easily run older versions of Photoshop.

1 upvote
Carsten Saager
By Carsten Saager (Oct 19, 2011)

I also think this is the algorithm - at least in the outline. A bit stupid to use an artificially blurred image in the presentation.

I do work in a similar area and artificial input allows to compare changes to the algorithms in a controlled manner.

And yes, this is a quite CPU intensive operation, it is o(N log N) because of the FFT that eventually does the heavy lifting

2 upvotes
samlew
By samlew (Oct 19, 2011)

I think what everyone is forgetting is that these are SNEAKS. Sneaks mean, "we're showing you something that we're developing and something that MIGHT be in a future product" (no promises).

So if the technology isn't completely done, or they show an example of what it might do when completed, I don't have a problem with that. They're not trying to sell it to me now. They're just showing what they're working on. And knowing Adobe, they'll get it perfected before another version of Photoshop is released (or whenever).

Making a huge deal about this is simply silly—it's a "sneak". But maybe it's just me. ;)

3 upvotes
Austin101
By Austin101 (Oct 19, 2011)

My research team have been out in the field gathering statistics and I can confirm

drum roll.....

it's just you

3 upvotes
vvitos
By vvitos (Oct 19, 2011)

....I cant believe how so many people are SO negative with things...
too much blablablablablabla....
adobe did what adobe did and they announced it by themselves...
they explained it is a common practice and I take their word for that...
adobe has given us photoshop a great tool for photographers all over the world..and if they say that they try to fix blur photos I don't care what path they follow as long as they give me such a major breakthrough...
I will stop deleting blur photos..maybe in the future adobe will manage to fix them for me...
until then...keep on trying guys! we are with you 100%!

2 upvotes
littleroot
By littleroot (Oct 19, 2011)

weasels

0 upvotes
Austin101
By Austin101 (Oct 19, 2011)

lol was that the best excuse they could come up with

the first photo of the blurred crowd doesn't look like real camera blur either.

in time with better sensors which collect a greater amount of detail this might be possible, some kind of short video exposure would hold enough detail to correct back to a sharp still image.

the logic here is that the software just pushes back the blur to where it should be because all the detail and information is in the blur....really?

big fail and busted

0 upvotes
AvanGarde
By AvanGarde (Oct 19, 2011)

The fact they used synthetic blur means most probably this feature is nowhere as good they promise but they scored some marketing points.Who cares about truth, truth is what they say it is.

I remeber they guy in the demo told they 'found' this 'blurry' photo of their friend.This is simplyl a lie, the photo was never blurred it was made to be blurred thats huge difference.The cheated and lied to deliver good marketing effect, because there is no way they couldnt not take a real blur photo.But real photo would most probably not impress audience so here you are.

1 upvote
Peter Rongsted
By Peter Rongsted (Oct 19, 2011)

It may just have been a small cheat - but they were cheating!

0 upvotes
bugbait
By bugbait (Oct 19, 2011)

Well, yep it was stupid beyond belief, just amazingly stupid they did that, and more than one person would need censure or dismissal. No that was not for comic effect. It was a cheat plain and simple or at least equals the same thing.

As for when such a thing would ship in PS*, I would imagine they would need specific highly classified data for each lens and or camera from each manufacturer. Even if a particular manufacturer was forth coming it wouldn't mean it would do as equally well as the other manufactures as they complied.

I think I will be buying another tripod or two before I see this in PS for most cameras.

0 upvotes
gl2k
By gl2k (Oct 19, 2011)

The technique itself is very promising.
Add "deblurring" to VR, OS, ... and at the end you might get 6 to 7 stops more.
Cam manufacturers will eventually license it and built it right into their P&S cams.

0 upvotes
firebr1ng3r
By firebr1ng3r (Oct 19, 2011)

i ain't even mad bro

0 upvotes
Debankur Mukherjee
By Debankur Mukherjee (Oct 19, 2011)

Why Adobe Why ? Was it too hard for you go go out and take a photograph at a slow shutter speed........this is not "research" you were demonstrating the possible future of Photoshop in front of few hundred people......

1 upvote
Tee1up
By Tee1up (Oct 19, 2011)

It is pretty hard not to be a little suspicious of the technology when they tell you we faked some of the results. Try and imagine me posting some of my Canon A620 photos here and then later telling everyone that 'some' were taken with a Leica M9.

1 upvote
Joseph S Wisniewski
By Joseph S Wisniewski (Oct 19, 2011)

I'm not particularly impressed by the "real" images that they used, either. Look at the deconvolution kernels, there's a "dwell", a time where the camera was steady enough to record a sharp image for part of the exposure time, so the movement adds a "ghost" to that, but there's still a sharp component.

Having run hundreds of images through similar algorithms (generally, I'm more interested in the analysis of the motion or of the point spread due to optical flaws, the actual deconvolution generally turns out to be a mess) I can tell you what rarities the images Adobe "cherry picked" for their demo are. If even those 1 in 100 flukes weren't dramatic enough for their demo, necessitating the totally faked image, that's a pretty good assessment of the general uselessness of this sort of thing.

And that's the "KEY point" you can "takeaway".

3 upvotes
Michael J Davis
By Michael J Davis (Oct 19, 2011)

That's a good point. But, as a technical person - but 'lay' in this area - I've been analysing my camera shake problems for a number of years. (Examining highlight trails of images between 1/50th & 1/4 sec)

Of course, with IS, the pattern changes (I think for the worse), but typical is a dwell on the subject, followed by a rapid move and restabilisation, often at right angles to the shift. That gives a sort of blobby-ended 'L' - I was amazed to see that as the decon. error in the first Adobe example. So I'm surprised that you find it untypical.

If, as I now assume, they pick up the main parameters from a 'dwell' and track the weaker image to another dwell, ISTM, that this should produce the best solution. I guess in this field 'perfection' ain't possible;-)

Mike

2 upvotes
ProDesignTools
By ProDesignTools (Oct 19, 2011)

The KEY point not to forget is that the most impressive examples - the first two images - were restored after actual camera shake.

http://blogs.adobe.com/photoshopdotcom/files/2011/10/Plaza.png

Everything else seems tangential to the takeaway.

2 upvotes
Kjartan Haugen
By Kjartan Haugen (Oct 19, 2011)

How do you know there where actual camera shake..? How can we be certain?

4 upvotes
jkrumm
By jkrumm (Oct 19, 2011)

First Olympus, now Adobe. Everything's a blur.

4 upvotes
Diopter
By Diopter (Oct 19, 2011)

withholding part of information about the scene is the most common way of achieving "artistic" results.
It is the market demand and the Photoshop always has been responding to that. What's the difference with the discussed case?

0 upvotes
bm bradley
By bm bradley (Oct 19, 2011)

I'd be impressed if adobe could manage to write a data base that didn't crash all day long... PS is a very buggy, crash prone app... in fact the only software I use that crashes several times a day... $.02

2 upvotes
wbsy
By wbsy (Oct 19, 2011)

Curious how they synthetically blur an image that looks like coming from the camera. Could be a nice feature too! I hope to see it work in both way in the final release.

1 upvote
rinkos
By rinkos (Oct 19, 2011)

that deblurring video is one amazing feat out of ADOBE....now when is it gonna be out..u better put it on ur upcoming CS6

0 upvotes
BJN
By BJN (Oct 19, 2011)

Wow, what a tempest in a teapot. If you watch the demo, the image in question is shown mostly for comic effect after the technology is show in detail on two larger images that are truly blurred by camera motion. Some marketing person helping to put the presentation together decided to add a little humor and thus the "synthetic" blur was born. If the algorithm can extract a motion formula to remove blur, it makes sense that the motion formula could be applied to a sharp image in a reverse of the process. Frankly, although it wasn't mentioned in the presentation, such a synthetic blur could be a useful tool. Motion blur isn't organic or realistic.

Adobe's technology works in two directions instead of one. You'd think this was some big coup on the part of pixel peeping web geeks, but it's just a very minor bit of unexplained showmanship on the part of Adobe marketing. Grow up internet, and shame on DPReview for jumping on the bandwagon.

6 upvotes
Richard Butler
By Richard Butler (Oct 19, 2011)

Shame on us? Really?

Adobe publicises a development that causes a lot of excitement (over 1M views on YouTube). Then somebody makes the completely correct assertion that all is not quite as it seems.

Having published a story about the video, we owe it to our readers to ask Adobe about the allegation. We then publish the clarification - which makes clear that the two images gaining all the applause were genuine but the suspicions about the final image were well founded.

I'm sorry, but in what sense should we be ashamed of trying to clear the matter up for our readers?

20 upvotes
cactusgeorge
By cactusgeorge (Oct 19, 2011)

BJN
We're just having a bit of fun. Don't you think it's funny for Adobe to photoshop a blur and pass if off as authentic so they can fix it in photoshop?

3 upvotes
huyzer
By huyzer (Oct 18, 2011)

Haha. Wow. Caught red handed. Nice work Adobe. Sheesh!

0 upvotes
Marcin 3M
By Marcin 3M (Oct 18, 2011)

I wonder if the shake trajectory was extracted from the image, or rather recorded by accelerometers. They stated that it was calculated, but this calculation is time-consuming. I would like to know the theory behind.

Comment edited 18 seconds after posting
0 upvotes
whtchocla7e
By whtchocla7e (Oct 19, 2011)

And who's going to reveal to you their trade secret?

0 upvotes
Joseph S Wisniewski
By Joseph S Wisniewski (Oct 19, 2011)

The blur (or "point spread") function is calculated by one of a family of algorithms known as "blind deconvolution". I generally use something known as "homomorphic deconvolution" or "cepstral analysis".

The calculation of the blur is not particularly time-consuming, because it's done on small "tiles" of the image, not much larger than the PSF, and you typically don't need to calculate it over the entire image.

The problem is that the inverse of the PSF is infinite in size, and even calculating a convolution kernel larger than the image itself causes artifacts. So, you're dealing with a 2 dimensional convolution of two "image sized" data sets. The number of computations is proportional to the square of the megapixel count, in the hundreds of trillions for a full sized image. Even using FFT techniques to accelerate it still leaves you with billions of calculations.

3 upvotes
Pdf Ninja
By Pdf Ninja (Oct 19, 2011)

It might not be necessary to FFT the whole image. They might be able to get away with FFTing just certain fragments of it. Once the kernel is known, of course the deconvolution must be run on the full resolution.

I haven't done deblurring, but have done deskew / auto-straightening of scanned images. First you find a usable area that you analyze. Or let's suppose you do Hough transform to find an angle of a black border. Instead of computing the sine waves at full resolution, you do a low resolution estimate first, then once you have the approximate angle, you compute it at higher resolution, but only around the estimated angle.

I'm just saying there might be ways to avoid having to FFT the entire image.

1 upvote
Pdf Ninja
By Pdf Ninja (Oct 19, 2011)

My bad. Convolution is actually algorithmically more complex than FFT. So yeah it's almost certainly impossible to do anything better than O(N * log N).

0 upvotes
bartg
By bartg (Oct 18, 2011)

So . . . Adobe has included and solved for the introduction of temporal distortion from rolling shutter effects? Because of this, the motion “profile” is not always static across the entire frame.

Good luck on that.

1 upvote
cactusgeorge
By cactusgeorge (Oct 18, 2011)

This to be a completely valid argument since there are no blurry photos online. Also, taking a blurry photo of Kevin just to maintain integrity of the demonstration seems like overkill.

1 upvote
gw5815
By gw5815 (Oct 18, 2011)

I was there. The most amazing part of the demo was the crowd picture and poster picture, and both of those were legit. The Kevin Lynch pic was just the comedy at the end.

The demonstration was crazy crazy awesome.

Greg

3 upvotes
joe6pack
By joe6pack (Oct 18, 2011)

If you look at the crowd picture, it is taken with good day light. You have to be pretty messed up to be ended up with motion blur.

http://blogs.adobe.com/photoshopdotcom/files/2011/10/Plaza-1024x511.png

0 upvotes
Caleido
By Caleido (Oct 19, 2011)

That crowd picture doesn't look like genuine, all day blur at all. It has sharp edges and detail in it, it just doesn't look right.

4 upvotes
Morris Sullivan
By Morris Sullivan (Oct 20, 2011)

With that much motion blur wouldn't you expect the moving people to have some blur as well even in the corrected photo?

2 upvotes
Total comments: 97
12