• Quick note - the problem with Youtube videos not embedding on the forum appears to have been fixed, thanks to ZiprHead. If you do still see problems let me know.

Adobe deblur technology - amazing

At first I thought it was too good to be true but after reading about the limitations in the article I'm much more enthusiastic. It's certainly believable within those limitations and I think it is a wonderful new tool!
 
I have some blurry pictures I will be saving instead of deleting because of this, for when I get a chance to use it.

Next step, deugly.
 
Just wanted to note the examples didn't come from the algorithm they'll be using. It looks more like a "Here's what we expect it will be able to do" example rather than a "Here's what happened when we ran this through" example.
 
Mister Earl said:
Just wanted to note the examples didn't come from the algorithm they'll be using. It looks more like a "Here's what we expect it will be able to do" example rather than a "Here's what happened when we ran this through" example.
Click to expand...


Are you sure about that? The text is somewhat ambiguous with the first set of images, but for this set, it does sound like they are actually using the prototype technology: "For example, you can see how well the prototype deblurs the text in the image below."

http://blogs.adobe.com/photoshopdotcom/files/2011/10/deblur-poster.png
 
If I'm reading between the lines correctly, this trick is very effective with motion blur, particularly if the motion is more-or-less constant during the exposure. It won't help much with defocus or jitter.
 
That's awesome! :cool:

They should go ahead and ship it now. Even if it doesn't work on every image yet, users could still salvage some images that otherwise would have to be junked.

[Picard]Computer, enhance.[/Picard] :D
 
They've had this on CSI for years.
"Zoom in.. sharpen... sharpen.. That's it!"
 
dasmiller said:
If I'm reading between the lines correctly, this trick is very effective with motion blur, particularly if the motion is more-or-less constant during the exposure.
Click to expand...
Constant seamless gradual motion probably kills the photo beyond remedy.

To me the top left photo looks more like a sudden shake that created maybe 3 pictures on top of each other, a bit like having three dia sheets of the same picture on top of each other, sligthly misaligned. So what the algorithm does is realign the three dia sheets to match each other as perfectly as possible. The difficult part is obviously exctracting the three different images from the blurred original image. But the key to success is the fact that you expect to find three _identical_ pictures from three positions here. It reduces the need for guessing, you just need to calculate which pixel colour in position X,Y is equal to the source pixel colour in three misaligned pixels whose position is known precisely, and which are affected by misaligned pixels of two different positions of the two other misaligned layers, whose exact position again is known precisely. Lotsa circular cross-references needed to process the entire image.
 
Last edited:
marplots said:
They've had this on CSI for years.
"Zoom in.. sharpen... sharpen.. That's it!"
Click to expand...

I've always wondered if this violates some principle of information theory. For that matter, does information theory give some upper limit on how badly something can be blurred and still recovered?
 
Offhand, it appears to me this new feature is the unsharp mask function on steroids.
 
Last edited:
phildonnia said:
I've always wondered if this violates some principle of information theory. For that matter, does information theory give some upper limit on how badly something can be blurred and still recovered?
Click to expand...

Depends. Does intuition matter?

Adobe Vs Captcha Vs Me, trying to figure out if I went to school with that guy in the Shroud of Turin.
 
hmm.. are the pictures being deblurred here actual blurry pictures or artificially blurred ones? Just curious.
 
phildonnia said:
I've always wondered if this violates some principle of information theory. For that matter, does information theory give some upper limit on how badly something can be blurred and still recovered?
Click to expand...

I wouldn't have thought so - the algorithm could be adding information as well as just extracting information.

I saw a very early prototype years ago of a car numberplate recognition system that seemed to be able to "deblur" and restore to pin-sharpness letters from a fuzzy photo of the number plate. What that system did was work out the shape of the letters from the high contrast regions and then painted on the actual font so it looked as if the characters were "restored" from very little information when that wasn't what was actually happening. (In the UK number plates have to use the same font, spacing, size of letters and so on so once you have worked out what the characters are its relatively easy to paint on the correct characters from a standard font.)
 
That is cool beyond cool. Can't wait for the plug-in.
All I need now is to get Hawking, Einstein and Newton around to show me how to use CS5.
 
I wonder if they developed this or are perhaps licensing the technology from that start-up that introduced the camera a few months ago where you can just take a picture and focus it later?
 
phildonnia said:
I've always wondered if this violates some principle of information theory. For that matter, does information theory give some upper limit on how badly something can be blurred and still recovered?
Click to expand...
Not on a TV drama.
 
welshdean said:
That is cool beyond cool. Can't wait for the plug-in.
All I need now is to get Hawking, Einstein and Newton around to show me how to use CS5.
Click to expand...

Even they can't figure that out.



The Central Scrutinizer said:
I wonder if they developed this or are perhaps licensing the technology from that start-up that introduced the camera a few months ago where you can just take a picture and focus it later?
Click to expand...


Odd... I was flashing back to that thread too.
 
You must log in or register to reply here.

Back
Top Bottom