Researchers develop algorithm to combat photo blur
Since it's unlikely that your hand will get any steadier with age, and we probably won't see optical image stabilization in cameraphones anytime soon, researchers are concentrating on ways to fix your crappy photos once they've already been captured. The latest salvo in the war against so-called hand motion blur comes from a team of computer scientists at MIT and the University of Toronto, who have developed an algorithm that can create a sharper picture by "estimating the distribution of a number of probable images" and coming up with a happy medium. Introduced at this year's Siggraph Conference in Boston, the algorithm could potentially be included in future versions of Adobe Photoshop -- which currently fights blur with a rather ineffective unsharp mask tool -- although it will do nothing for blurring caused by moving objects or improperly-focused shots. Unfortunately, it sounds like this product is still at least a year away from commercial release, so tripods and nerve-steadying Pentazemin are still your best bets for the time being.
[Thanks, Alex]
[Thanks, Alex]



















Hello,
I’ve been looking for months for a tool that actually does image resizing by maintaining quality of the picture.
I’ve tried Photoshop and various others but I got blurry contours and I’ve lost the sharpness of the image.
I’ve stepped over a new tool that I’ve found at http://reshade.com which satisfied me completely so I thought that there must me more like me out there, regular people who were amazed by this video but didn’t quite know what to do with it, who just wanted to use the feature in it.
Actually, in PS CS2, there's a great filter for cleaning up virtually any kind of blur that you can throw at it, whether caused by lens zoom, motion blur, radial blur, etc. It's worked swimmingly for me.
What filter is that Ryan?
maybe engadget can use this to clear up some of their horrible leaked pictures that always seem to be blurred the same way.
Convolution wins again!
How's the anti-ugly filter coming along?
They couldn't find a better model?
apologies for the correction and for yelling, but it's properly spelled "SIGGRAPH".
Wiki: http://wiki.siggraph.org/wiki/Main_Page
if you take several images in succession, you can use a technique called "stacking" to improve apparent image quality. It's not good for scientific accuracy, necessarily, because it's interpolating the image, but it's still visually useful. Many amateur astronomers use this technique when they do digital imaging, to help get around transient atmospheric effects, etc.
Quick!, Use it on the pictures of HTC's Excalibur!
um...Unsharp Mask isn't really a tool to be used for correcting motion blur. Typically you use it on an images that is already very sharp to enhance a certain area - say the iris of someone's eye or the strands of their hair. I don't know of an easy way to combat motion blur (I wish Ryan would enlighten us!) but if you use the wrong tool for any job, you can't really blame the tool for being ineffective, can you?
@ Alex
Smart Sharpen. :|
Of course, it helps to actually take decent pictures to begin with.
Stacking doesn't work for photos of people... Unless they are frozen, and you're using a tripod.
I predict that millions of photos of ghosts, bigfoot, Jesus, Elvis, and UFOs will suddenly prove the existance of Underpants Gnomes.
This is likely a "blind deconvolution" technique. Convolution is the mathematical process that combines and mixes signals. In this case, the original signal can be interpreted as a "mixture" of two photographs exposed at slightly differing instances in time to the CCD. These "two" images are convolved to form one - the blurry image. Bilnd deconvolution works backward to solve for the "mixing" factor of the two images (the "blind" part, because you don't know the mixing factor), and then deconvolving means UNapplying the mixing factor to recover the original, hypothetical image.
It is essentially a reverse engineering approach, and I would be very surprised if Photoshop's sharpen effect performs this very complex mathematical task. Blind deconvolution is traditionally very slow and very computationally intensive, and could take minutes to perform on just one picture on today's CPUS. But anything is possible :)
uh, right. it's ineffective to just about every serious photographer who uses it on a daily basis to improve their images. it might be ineffective to some two bit moron who can't take a picture with his terrible quality cell phone camera or to someone expecting a sharp and noise free image taken with some f/5.6 point and shoot inside a museum or at night, but to a real photographer, unsharp mask is an amazing tool.
I thought that for certain types of motion-induced blur, particuarly when the motion is at constant velocity and transverse to the image plane, some FFT techniques work to effectively de-blur the image. It's been a long time, but I think you can process pictures in Matlab this way.
On top of Smart Sharpen, Photoshop already has a filter called Motion Blur (it's under Filters/Blur). But it's not great; in fact, I never could figure out how to use it.
Maybe the camera just did not like the model because of her lack of "appeal." And yes, as the poster above said, where is the anti-ugly filter??
@ John Doe
Motion Blur is to blur the image to look like the picture is in motion (or the camera is in motion), haha, not to un-blur it.
Interesting. I've been using this PS tool called FocusMagic to fix motion blurs and has worked wonderfully for me. Contrary to the software's name, it does not fix *focus* but actually the *motion* blurs. But it needs some manual fiddling with vector -direction and degree (pixels) of blur and doesn't work on complex ones, you know like lot of shake or circular movements.
If this MIT thing is fully automatic and can be done realtime in-camera it's great!
On second thoughts, why can't camera makers include a motion detection sensor in camera (has to be a tweaked gyroscope or something that even detects radial movements... well any sort of movement!) Based on camera shake, the A/D does motion vector sampling every few milliseconds and feeds that data to the on-chip software which now has all the information it needs: shutter speed, aperture, focus & motion vector history. Some complex mathematical treatment, and we should have our blur-free picture.
My 2(0) cents :)
http://www.focusmagic.com/examplemotionblur.htm
Best way to combat photo-blur is take a better picture. Use a tripod if you hands are that unsteady.
I would like to see how the algorithm works on images that are not caused by a shaky photographer; that is, the model themselves actually moved (and possibly changed expression, etcetera). Or where the shake is not so linear.
But bumping a camera during a long exposure, while the subject is relatively still, blurs everything in the image more or less equally (save for parallax differences), and this might be a nice addition to the rest of photoshops filters. I don't know if I would pay for it as a plug in.
I hate to disagree.. But my k800i has image stabilisation...
Shame that work is a complete ripoff of David MacKay's at the University of Cambridge. The ignorance of the graphics community strikes again..
Another program that can handle motion blur is unshake, with very good results...
I use FocusMagic as well... it fixes motion blur and out of focus photos both. It's basically a manual version of the software in this article -- you must tell it which direction the camera moved and how much (as determined by zooming in and analyzing a point of light etc) but once you have this information it does a fairly good job of putting the image back together.
"uh, right. it's ineffective to just about every serious photographer who uses it on a daily basis to improve their images. it might be ineffective to some two bit moron who can't take a picture with his terrible quality cell phone camera or to someone expecting a sharp and noise free image taken with some f/5.6 point and shoot inside a museum or at night, but to a real photographer, unsharp mask is an amazing tool."
USM is not for correcting blur, it's for sharpening already sharp photos or photos that have been resized.
"Interesting. I've been using this PS tool called FocusMagic to fix motion blurs and has worked wonderfully for me. Contrary to the software's name, it does not fix *focus* but actually the *motion* blurs. But it needs some manual fiddling with vector -direction and degree (pixels) of blur and doesn't work on complex ones, you know like lot of shake or circular movements."
Focus Magic can work with both out of focus blur and motion blur, although I rarely have good luck fixing motion blur. Photoshop's new smart sharpen has technology that works the same way as Focus Magic. Except it's a lot faster and allows fractional radii. For instance, in Focus Magic, you can choose a radius of 1 or 2 or 3. In Smart Sharpen you could use .6 or 1.3.
Photoshop's Unsharp mask consists of a blurred version of the image, which is subtracted from the original image. You can create the exact same effect with the blur tool & layer composition options. For a motion blured image, you could instead use the motion blur effect to generate the subtraction image.
Here is a simple explanation of how the unsharp mask works:
http://www.cambridgeincolour.com/tutorials/unsharp-mask.htm
Andreas, thanks for that. Very interesting.
"Photoshop's new smart sharpen has technology that works the same way as Focus Magic. Except it's a lot faster and allows fractional radii. For instance, in Focus Magic, you can choose a radius of 1 or 2 or 3. In Smart Sharpen you could use .6 or 1.3."
Nope. Who told you that? SS uses intelligent comparison of gaussian vs original versions with variable radii. I'm a developer myself so although I can't reveal much here,the exact focus/motion blur feature u are talking about should be there in Adobe Lightroom final. Until that time, do test drive the beta 3 mate!
I'm at SIGGRAPH and saw the presentation - and it was an excellent presentation. This algorithm is anything but a rip off of David MacKay's work. The project page and specifics can be found at :
http://people.csail.mit.edu/fergus/research/deblur.html
I'd like to note that some cameraphones (including my FOMA D902iS) do already have handshake/blur reduction, which work incredibly well. I've tested it in the 4MP setting purposefully swinging the camera and get a perfectly clear shot.
So at this rate it'll be in the west by atleast 2010.
We have an interactive approach to applying Fourier inverse filters to tackle real world blurs. See our website for many examples. The hard part in deblurring is to measure the blur (point spread function ) exactly. Our method lets you see trial restorations interactively, while moulding the psf, to get accurate results. We offer a service on the internet... Try us!