10
\$\begingroup\$

I have an image which I have modified through Picasa ( using Soft Focus - essentially I am making the faces blurred ).

Question is : If I upload that photo to Facebook, and a user downloads it from Facebook, is there any way they can modify the photo so that the face is in focus again?

PS: I am not doing anything mischievous - my subject is against a brick wall holding a Nikon camera, and I just want to focus on the camera strap so I've blurred everything else.

\$\endgroup\$
2
  • \$\begingroup\$ Hope you don't mind my edit. I don't think this really has anything to do with Facebook. It's more to do with how you've blurred the original (and how much) - and whether it's beyond the point that someone could bring it into enough focus to recognise the subject? \$\endgroup\$
    – MikeW
    Feb 24, 2013 at 9:18
  • \$\begingroup\$ @MikeW not at all - I have brought it totally out of focus using Soft Focus - so the Camera is below waist line of the subject and every thing else is blurred \$\endgroup\$ Feb 24, 2013 at 9:20

6 Answers 6

8
\$\begingroup\$

There is a technique called deconvolution which can, to some extent, recover distorted or blurred image detail.

Topaz labs have a product called InFocus which uses this sort of technique. It can sharpen certain images, but if you have blurred your faces substantially, I believe it would be beyond the ability of any tool to recover. Maybe on CSI :P

\$\endgroup\$
3
  • \$\begingroup\$ you meant CSI the TV Show or some tool ? \$\endgroup\$ Feb 24, 2013 at 9:33
  • 1
    \$\begingroup\$ Sorry, meant the TV show, as a joke :) \$\endgroup\$
    – MikeW
    Feb 24, 2013 at 9:33
  • 1
    \$\begingroup\$ ha ha :D .. sorry i dont have any idea on Photo edit software ..so making sure .. I will send it to 24 team too :D \$\endgroup\$ Feb 24, 2013 at 9:35
6
\$\begingroup\$

If you're doing this for security or privacy, the only sure way is to really mask out the faces with solid color. A sufficient amount of blur will be destructive, to the point where reconstruction techniques like deconvolution won't be effective — but if the image will be subject to scrutiny, it might be hard to judge how much is needed to be safe. Which means that the safe thing to do is use a solid overlay rather than something which is inherently based on the underlying data.

Certainly don't use a filter like mosaic or distortion; it's sometimes amazing how well these can be reversed.

\$\endgroup\$
6
\$\begingroup\$

I work in data security and follow some data security news, images with data hidden by bluring and other image manipulation are "broken" all the time.

Digital filters in software work by taking the image data, doing some math on it and producing the modified image.

Unlike images that are shot out of focus in camera, with digitally blured images we know exactly the math formula the software used - and often we can reverse the formula and get the original image back (sometimes with some quality loss)

I have no idea what Picasa soft focus uses and I don't know if it's reversible, also I'm not going to analyze it, treat this answer as a general warning not as a review of Picasa'a soft focus filter's security

Even if we can't un-blur the image often it's possible to recognize the blured person/object/text/whatever by bluring sections from other images and comparing them (because if we take two similar image and apply the same math to both we get two similar changed images).

So, your bluring method will stop the casual surfer but not anyone who really want to get the data, never use it to hide really sensitive data (especially not text or numbers), be careful and if you want to hide something just paint over it with a solid color (and check it's also hidden in thumbnails, preview images, undo histories and such).

\$\endgroup\$
1
\$\begingroup\$

No, the detail is lost and a JPEG does not contain 'history' to allow the blurring to be undone.

Try it: save an image with the faces blurred as a JPEG, then re-open it and try to recover the detail.

\$\endgroup\$
2
  • \$\begingroup\$ wondering if this is possible in Advanced tools like Photoshop ? \$\endgroup\$ Feb 24, 2013 at 9:23
  • \$\begingroup\$ Hi Tony, you've got some great answers here on Photo SE, but can I remind you that you are not meant to sign your answers/questions with your name, your user card does this for you. \$\endgroup\$
    – bazite
    Feb 24, 2013 at 16:27
1
\$\begingroup\$

The short answer is a loud NO. What you are asking is equvivalent with me giving you a number, lets say, 10000, and I ask you what summation this number is a result of, lets say 5000 + 5000. There is 4999 other possibilities, thus there is no way you can calculate this, but are limited to wild guesses. A blurry image is essentially polluted by light from adjacent points in the object to adjacent pixels in the sensor. Thus every point in the image is a sum of many points in the object.

Though, there is not zero information about the object in a blurry image. For instanse, if you had another image of the same object (face) you can run different digital filters to get answer to this question: could this blurry image be of the same person I've got another, sharp image? This is basically the technique being used by people who claim they can reconstruct lost information in an image. They assume stuff about the object.

If the answer is yes, it wouldn't prove it is the same face because a blurry image would fit almost unlimited amount of faces, depending how blurry the image is.

This is the reason why photographers, good ones I might add, stress the point: you can never recover lost information in post production. You can only post produce a new image from an image you dislike so that it looks good according to certain emotional driven standards, but it has nothing to do with the object.

\$\endgroup\$
5
  • 1
    \$\begingroup\$ The question is about software generated blur. Your point is essentially valid if the OP were trying to correct a badly focused image. you can never recover lost information in post production => as Nir said in his answer, sometimes you can recover information if you know the algorithm used. Moreover, concerning motion blur, some tools give good results : cse.cuhk.edu.hk/~leojia/projects/motion_deblurring/index.html \$\endgroup\$
    – Olivier
    Aug 5, 2016 at 13:32
  • 1
    \$\begingroup\$ This is not true for the case of badly focused images either, because the defocusing process can be formally described using an unknown model that nevertheless only contains a small amount of information (a few bits) while the picture contains a huge amount of information. So, out of focus images can be de-blurred using deconvolution methods, even little is known about the blurring, e.g. using so-called blind deconvolution methods. Or you can just consider what should be sharp edges in the pictures that are now blurred and solve for the point spread function. \$\endgroup\$ Aug 5, 2016 at 18:11
  • \$\begingroup\$ Weather it is software generated or analog spread of light is not relevant. Even if you knew the software function that generated the blurry image there is no way you can calculate 10000 back to 5000 + 5000. This goes for each and every pixel in the image. This guessing game has almost infinite amount of conbinations. The link you are refering to shows so little understanding of the problem that he has not even given his readers the original picture. This is a far more realitic take: goo.gl/5otll1 \$\endgroup\$
    – user55517
    Aug 5, 2016 at 22:39
  • \$\begingroup\$ I should have added in my anser that blurryness in context of identification one must remember that it is enough information to distinct one face from most other faces if you got as few points as hundred, each with a dynamic range of as little as six bits. This should be taken into consideration when choosing how much blurryness you should use not to be identified. Because of how little information that is required may make people to believe it is possible to reconstruct a blurry image, but it is always a completely new picture where not a single pixel has the correct value. \$\endgroup\$
    – user55517
    Aug 5, 2016 at 22:55
  • \$\begingroup\$ @user55517 You are wrong. As an oversimplified example, create "blur" by adding the two adjacent pixels (when existing) in a row, getting [4,7,2,8,15,4] as a result. The original can only be [16,-12,3,11,-6,10]. Indeed, if you are getting a CT or MRT scan, lots of point images resulting from "blurring" an entire cross sectional line are resolved into a sectional image. If the blurring MFT is constant or slow-changing, much of it can be inverted while amplifying image and quantisation noise (the tradeoff is optimised with Wiener filtering). Movies additionally allow time integration. \$\endgroup\$
    – user98068
    May 19, 2021 at 12:38
0
\$\begingroup\$

If the information has been lost the information has been lost, BUT a skilled user could easily enough add in "data" from other face images and reconstruct your image so that it looks like the face belongs.

\$\endgroup\$

Not the answer you're looking for? Browse other questions tagged or ask your own question.