Incredible CSI-Like 'Enhance!' Software Fixes Unfocused Photos

Remember all those movies and TV series in which a FBI agent turned a completely blurred photo into a focused image by clicking one button? Remember how you sneered? Well, soon you will not be able to laugh at it anymore. This app lets you deblur any heavily blurred image just like that.

Even while imperfect, the results are truly astonishing.

Created by Vladimir Yuzhikov -- a software engineer specialised in image and signal processing -- the SmartDeblur app is truly impressive. According to him, the "restoration of distorted images is one of the most interesting and important problems of image processing -- from the theoretical, as well as from the practical point of view." Even while the maths for doing it started to develop about 70 years ago, he says, it's only now that we are seeing the results of it.

Yuzhikov worked on two common cases: blurring due to incorrect focus and blurring due to movement. "Each of you knows very well," he points out, they "are very difficult to repair". In fact, the examples used as samples by Yuzhikov are impossible to fix using current commercial technology.

Magic processing

He started his work with his own picture of Venice. He applied a synthetic blurring filter on it and then he applied his software to it. It worked, as you can see it above. And while it didn't revert to its original state, the results were impressive.

Many people think that blurring is an irreversible operation and the information in this case is lost for good, because each pixel turns into a spot, everything mixes up, and in case of a big blur radius we will get a flat colour all over the image. But it is not quite true -- all the information just becomes redistributed in accordance with some rules and can be definitely restored with certain assumptions.

But, of course, artificially blurred images are not the real problem. The real problem is real-world fuzzy images caused by poor lens focusing or unwanted motion blur. The results, however, were not as perfect but equally impressive.

Take a look at his examples:

Example of blurred image because of poor focus; taken with a Canon EOS 500D camera using manual focus.

Yuzhikov's app result.

Example of blurred image because of motion.

Yuzhikov's app result.

Example of out of focus image of text (left) and fixed image (right).

The program.

While the images have artefacts, it's quite amazing to see the level of detail that Yuzhikov's program can obtain from these blurred images, like window panes on windows that were just muddy spots. I'm sure that we will soon see this and much better focusing filters in future software. And with the computational power of smartphones increasing by the day -- especially their graphic processing units, which are ideal for the maths required in this -- it will not be surprised to see this feature built in the camera software of future iOS, Android and Windows phones.

You can heredownload and try it (it's Windows-only for now). [Yuzhikov -- Thanks Ángel Jiménez!]



    It's pretty impressive. Deconvolution is awesome.

      Impressive yes, but I am wondering how accurate it is . Also, would the results be admissible as evidence in a court of law?

        It would be easy enough to validate by comparing "enhanced" photos to the subject of the original photo in a prospective, blinded manner when applied to photographed text, for example.

    Dont worry, Lytro will get rid of him

      lol hopefully lytro will bring SLR quality pic soon, but that means... everyone can be a photographer lol

    So this is the same thing they used on iOS Maps, yeah...?

      ^^ useless comment, got nothing to do with Apple

    The CSI model requires "zoom and enhance", not just removing blur. Lots of artefacts, but good if want to read a number plate.

    the photoshop deblur demos look promising. wonder when that will be ready for public consumption?

    It's not blurry photos in CSI that make me sneer, it is when they take 12 grainy pixels from an NTSC video and turn it into a legible number plate.

      You still need a decent quality picture to start with. Not zoomed in CCTV footage.

    I am sorry, but whats the big deal? At the point this software is at, it looks to me like the pics are made worse, not better... Its a good idea, but the software is not there yet. Not by a long shot.

      What's the big deal? This:
      The text in that image was UNREADABLE before processing... nuff said.

      It would be good for adding detail to , say, surveillance video which is notoriously poor. However, I doubt it would stand up in court, it may be useful for things like identifying number plates or tattoos etc.

      Last edited 22/10/12 10:31 am

      the big deal, I think is best shown in the text image above. previously, you would have to take another photo and hope for better results. with this, you can just... fix it... and make the text legible. whether or not it becomes a "good" photo is irrelevant, making completely blurred text legible is amazing.

    I'm more bothered by flipping it round the vertical then enhance!

    Its a nice algorithm, but relies on several assumptions regarding the original pixels. Sharpening the image up can add artifacts that are not present in the original image, so its not perfect.
    Other applications could include up-scaling low resolution images or video using 2-pass filtering (one to interpolate and another to "enhance")

    can it beat the face and license plate blurring technology that is applied to all google street view images?

    maybe now the scenes from (insert generic cop show) where they can focus onto a ridiculously small detail and can make it clear be more realistic!

    You mean so they can see who took the photograph? :-)

    All I can think of is Super Troopers..

    Enhance! *mash keyboard* Enhance! *mash keyboard* Enhance!


    This could really change the way police work through surveillance tapes!

    Super Troopers GO!

    Here an image deconvolution application in App store Photo Doctor

Join the discussion!