Page Content | Events
Other IMCs
Leeds Bradford IMC | UK IMC | Editorial Guidelines | Mission Statement | About Us | Contact | Help | Support Us

(((i))) Leeds Bradford

Read It. Write It. DO IT!

Blurred faces in photos, SOS: Stop blurring now!

Demi (machine vision student) | 13.05.2009 00:53 | Technology

SOS the cops know how to see your face even though it might be blurred in an online photo, but they can also infer useful information from your skin colour or pair of jeans!

Recognition of blurred faces using Local Phase Quantization (2008) at

* In this paper, recognition of blurred faces using the recently introduced Local Phase Quantization (LPQ) operator is proposed. LPQ is based on quantizing the Fourier transform phase in local neighborhoods. The phase can be shown to be a blur invariant property under certain commonly fulfilled conditions. In face image analysis, histograms of LPQ labels computed within local regions are used as a face descriptor similarly to the widely used Local Binary Pattern (LBP) methodology for face image description. The experimental results on CMU PIE and FRGC 1.0.4 datasets show that the LPQ descriptor is highly tolerant to blur but still very descriptive outperforming LBP both with blurred and sharp images.

Blurred face recognition via a hybrid network architecture (2000) at

* We introduce a hybrid recognition/reconstruction architecture that is suitable for recognition of images degraded by various forms of blur. This architecture includes an ensemble of feedforward networks each of which is constrained to reconstruct the inputs in addition to performing classification. The strength of the constraints is controlled by a regularization parameter. Networks are trained on original as well as Gaussian-blurred images, so as to achieve higher robustness to different blur operators. Face recognition is used to demonstrate the proposed method and results are compared to those of classical unconstrained feedforward architectures. In addition, the effect of state-of-the-art restoration methods is demonstrated and it is shown that image restoration with the proposed hybrid architecture leads to the best and most robust results under various forms of blur

When you hide a face by the means of a blur filter then your image processing software simply utilises an algorithm (a specific sequence of steps of calculations) in order to modify some pixels by replacing their colour with the colour of another pixel.

The image processing experts have now found a way to recognise faces from blurred photos, inverting (in practice) the sequence of the algorithm (by performing the same steps and calculations in the reverse order). This explanation is not fully correct in the scientific sense, but it is the only way to explain it to you in simple terms without resorting to complex mathematics (I study machine vision doing an MRes in the South East, and while working on my dissertation I freaked out after I stumbled upon these articles!) I just hope you can appreciate the danger!!

What this means is that when you upload a photo with blurred faces, the cops can simply process your photo with a forensics program and reveal your faces without the blurring effect! It's as simple as 1-2-3 or A-B-C!

The good news though is that the solution is even simpler: you just refrain from using the blurring filter and you paint the whole face with your image processing software's pen by only using one simple colour (black or white), but NEVER the skin's colour because if you copy it then it MAY prove useful for analysis! see this:

Contribution of color to face recognition (2002) at

* One of the key challenges in face perception lies in determining how different facial
* attributes contribute to judgments of identity. In this study, we focus on the role of color cues.
* Although color appears to be a salient attribute of faces, past research has suggested that it
* confers little recognition advantage for identifying people. Here we report experimental results
* suggesting that color cues do play a role in face recognition and their contribution becomes
* evident when shape cues are degraded. Under such conditions, recognition performance with
* color images is significantly better than that with gray-scale images. Our experimental results
* also indicate that the contribution of color may lie not so much in providing diagnostic cues to
* identity as in aiding low-level image-analysis processes such as segmentation.

By replacing all the face's pixels with another colour all the information is lost for good (except if you use an image format that keeps undo levels or other metadata) and no one can recover it (not even their all-powerful god).

But I have more for you... there is also a program that can recognise the same person in two different photos as long as she or he wears the same pair of jeans!! have a look at this:

A Garment in the Dock; or, How the FBI Illuminated the Prehistory of A Pair of Denim Jeans (1998-2004) at

* This article looks at research carried out at the FBI Laboratory’s Special Photographic Unit in the identification of denim trousers from bank surveillance film. This research, which was published in 1998, showed that despite the ubiquity of jeans, each pair has individual identifying characteristics caused by the manufacturing process and by wear, and that these might be used as evidence in the identification of criminal suspects. What the FBI research also inadvertently illuminated was an otherwise hidden relationship between garment, maker and wearer, in an effective - if accidental - reversal of commodity fetishism.

Enlighten your comrades, translate it, and remember: knowledge is power! Freedom to everyone!

Demi (machine vision student)


Hide the following 25 comments

Dodgy advice -

13.05.2009 07:18

I'd suggest that you don't follow this advice for the reason that this person (who is very keen to impress us with her qualifiocations and knowledge, so we don't question her) seems really keen to get us to stop blurring faces just because it 'doesn't work'. I could see why she would be so keen if it activley helped the police identify us, but as the article claims that it makes little difference either way, i'm going to carry on blurring faces in photos - whats the harm?

(A) Sab x


13.05.2009 07:58

better is covering face with black pitch in windows paint


Think about it...

13.05.2009 08:21

The nonce that got picked up in Cambodia after poilce 'un blured' his face is a good example of unbluring.

If you blur the face using a blur tool then it can be undone. If you paint over the face and there is no metadata you ca not get the old picture back.

Paint then blur if you quite fancy the asthetics of bluring. I prefer the old Stalin esk Black Pen look personly..

James Smith


13.05.2009 08:23

You aren't being advised that covering up faces is a waste of time, you're being told that simply "blurring" them is not good enough.

Instead, paint over them with a solid colour.

Not much scope for disinfo there.



13.05.2009 08:56

Note that many image files created by digital cameras contain a separate hidden image which is used as a "thumbnail", for example when displaying image previews either on the camera itself or in a folder preview view on various operating systems. This image is usually the original unedited photo so could contain an unmodified picture of a person's face.

There are various free applications available on the web for JPEG metadata removal.



13.05.2009 09:09

blur, pixlate, blur

in the know

"and there is no metadata"

13.05.2009 09:50

That is a vital point that will be over the heads of most readers.

What it means is once you have your photograph in Photoshop/Gimp/ Paintshop /whatever, once you have obscured the faces by cutting or painting, then you have to copy the entire image and paste it as a new image. I imagine (hopefully) that IM strip metadata but other sites won't, so it is best practice imao.



13.05.2009 09:55

I have no idea if 'Guassian blur' can be reversed to any great degree. Never heard of the such and perhaps if you are good at advanced maths this will indicate whether the input number can be backtraced:

But given the lack of any 'unGuass' plugins, I suspect not... so the post may well be hoax.

The poster above got it wrong. The English paedo was caught because he used the SWIRL function, which moves data in a set pattern. Google "mr swirl" for more.

One thing is for sure, as another poster remarked, using a solid colour and a "paintbrush" will be irreversible... BUT only so long as the image is rendered down or saved to a jpeg. If you have a multilayered file (jpeg is never layered, so you will be safe using that format) the 'blacking/whiting out' can simply be switched off. So if you need the image in a non-jpg format, and aren't sure what I mean by 'render' or 'layer', then get a trusted person who does know to do it!

Mystery Photshopper


13.05.2009 10:11

Yup, earlier comments retracted - just skim read the article before work. Thanks for the advice.

(A) Sab x

What's gone is gone: Some more thoughts

13.05.2009 10:14

The IEEE abstracts in the article are vague.

In the first it doesn't specify what it means by blur (be it Guassian or motion) and in the second, despite the fact that Guass is mentioned it doesn't mention any tolerances of blur extent...

One thing it cannot be true, is that an algorithm could retrieve data that has been lost through the degradation of reiteration. In other words, if something is Gaussed to the degree that the image becomes one homogeneous tone, there can be no going back.

But if the Gauss has only lost 10% of the detail, then logically only less than 10% of the original detail could be accurately predicted by "forensics". And if the processes is predictive and thus less than 100% accurate, that means there are to some degree (likely increasing in inaccuracy in relation to the severity of blur) be artefacts introduced (as there are in exxisting sharpening tools, or in audio denoising algorithms. )

But basically the common sense rule applies: if the data is lost there is no brining it back... well not least till some brainiac comes up with some spiffy way of extracting the data from incidental environmental sources effected by the primary source (reflection etc.)

So the original post definitely smells of shit, if nothing else by its lack of caveats in line with what it is citing.

But again, one things is right: a block of black or white can't be reversed so long as teh image is "flattened" (rendered down... single layered...)

Kind of reminds me too of the J7 Truth idiots that started claiming the suicide bombers last will & testament videos were fakes, citing some obscure piece of animation software. Turns out the software was chiefly causing a stir because of it's voice formants to facial expression animation grammar, and was visually about as convincing as a sock with with two buttons on it in terms of being natural.

Mystery Photshopper


13.05.2009 10:29

What does you even mean by metadata? File headers or EXIF or both?

Exif will surely tell cops nothing much more than what camera was used and at what settings:

And surely any incriminating data within the file header will only be something that can identify the originating computer and would therefore be replicated by creating an new file on the same machine???

I'm not even sure if any image formats store non-image related data within the file (like MS Office files do)?

I can't start to imagine what either the EXIF or the file header could add to the forensic process at all (other than saving a minor step being needed) let alone being relevant to Guass?

The EXIF info can probably be all reverse engineered anyway.

Could you be more specific? Because it is news to me!

Mystery Photshopper


13.05.2009 10:37

Could you point to more technical sources/data on that.

I had hitherto believed the thumbnails were produced on an adhoc basis within the camera or within the OS.

This vague reference is the best I could find on a busy day:


Mystery Photshopper

Jeans and other identifiers

13.05.2009 10:38

I think this part of the article is interesting. For instance, I have recognised people I know in Google Street Maps who have their faces blurred from their general clothing, though I couldn't prove that in court. It suggests security services can identify people from even obscure details.

"This article looks at research carried out at the FBI Laboratory’s Special Photographic Unit in the identification of denim trousers from bank surveillance film. This research, which was published in 1998, showed that despite the ubiquity of jeans, each pair has individual identifying characteristics caused by the manufacturing process and by wear, and that these might be used as evidence in the identification of criminal suspects."

That doesn't apply to properly sized images on Indymedia, and to suggest it does is to encourage paranoia that would harshly limit any photographs here as no UK jury would convict on this supposition. It does however suggest black and white photographs are safer, less identifiable for clothes and such.


Remaining anonymous

13.05.2009 10:54

If you want to remain anonymous the best thing to do it make sure you aren't photographed in a way that means you may get into trouble or even at all.

Check out this article for tips on remaining anonymous

It's just as likely that if you are photographed by indymedia with identifiable features or clothing you have also been photographed by the cops or mainstream media who wont blur anything.


Learn somehting new every day...

13.05.2009 11:04

So, many digital cameras now embed thumbnails within the files and the thumbnails are not updated when people change the content:

And how to strip them out:

So the rule is:

When you edit out faces: strip out the thumbnails!

Mystery Photoshopper

super anon

13.05.2009 11:12

loads of nonces get caught when their stuff gets de-blurred.
but as mentioned before, it shouldn't affect us, as we should be masked up. no matter if we're being naughty or not.
but if their are none masked up people in a photo, I would print out a photo, put a black piece of paper (or marker pen) over the faces, then take a photo of my photo. easy as poo.
sometimes simple is the best, because they write the software we use, (unless your using gimp)
their will always be ways of uncovering people, and for all the bad the police do, I'm not gonna kick up a fuss about anti-peado measures.
mask-up and don't get caught out. end of

eye see you!

The original post is correct

13.05.2009 11:48

I've seen this debluring done before by computer sciences postgraduates when I was in Berlin for a while, it most definitely exists and the police do use it.

As others have said the best method is ruddy great big black blob on the face to cover it, you should then print screen it to a new image to strip some of the attached data, and then save it in a non multilayer format.

I thought everyone already knew that blurring was ineffective... apparently not. Debluring techniques have been used for years to catch paedophiles and rapists.

Stay safe

This article in Greek

13.05.2009 12:18


Stay safe

13.05.2009 15:37

Then perhaps you can give us a better explanation. Namely a run down of the technology and its limitations... Are you sure than Gaussian blur has been circumvented? I know the famous swirl case, but am not aware of any notable blur cases.

Just saying "it is correct" and adding yet more anecdotes doesn't really help anyone.

Mystery Photoshopper

photographer scum

13.05.2009 16:09

Black over the face, then fill with a flesh coloured tone and then blur that - it'll give you the blurred look and will waste them some time.

The thing is though, there are hundreds of other ways for the state to identify you, even without seeing your face.

I see little point in photographing activists whilst out. leave that to the fitscum.


Digital cameras have finger pritnts too...

13.05.2009 17:37

While removing exif and meta the photographer also needs to be aware that digital cameras have their own unique fingerprint

so we are back to the usual advice "if you are doing anything dodgy don't be photographed doing it and if some one is doing something dodgy don't photograph them" and if you do don't put it on the web


meta data goes on photoshop ?

13.05.2009 19:30

I beleive that as soon as an image is altered in Photoshop the Metadata dissapears automatically.

Can someone confirm this please.

.cold stomper

Cut n paste

13.05.2009 20:06

Dunno, but cut'n'paste anyway. A clean paste to a single layer jpg or png is fine, simplest solution imao. Don't be too paranoid about metadata unless you are the sort of person who gives your real name and address while installing programs. In which case you are probably a card carrying capitalist who doesn't have to worry about anything.

Cameras and printers have built in identifiers, search the archive here. Truth is this level of detail has never been brought to court except in paedo cases. In which case, stop abusing kids or be afraid of more than the police.

Ex Adobe Danny


14.05.2009 14:43

As already mentioned it's pixelation that must be used and not blurring.

Unbluring is nothing new, it is just difficult and cpu intensive.

And of course the easiest blurs to unblur are software generated blurs for obvious reason.


If two different faces blur to the same image, it can't be undone.

15.05.2009 15:58

If two different faces blur to the same image, it obviously can't be undone. That is basic logic.

The people who have been caught out (mostly paedophiles) have just used something to swirl or twist the image around. In that case you can just twist it in the opposite direction and undo it.

But in general a blurring is a one way process. Maybe statistical analysis of possible originals of a blurred image might be of some use in picking out which originals look like real faces. That's the route I would go down if I was trying to do this, anyway. It wouldn't work if a face was blurred multiple times with different methods though.

Probably best to blur the whole image slightly and pixelate faces as well, like some other people have said. The more detail the police can get from it, the better for them.

Obviously avoid posting images with people in them at all without their permission. But sometimes you may want to e.g. ALF activists posing with liberated animals, etc.

I read recently some paedophile got identified because part of his thumb was in an image and the thumbnail was slightly unusual.


Write It
Publish your news

Do you need help with publishing?


The Common Place - Leeds The 1 in 12 club - Bradford The Trades Club - Hebden Bridge


South Coast

Other UK IMCs
Bristol/South West
Northern Indymedia


satellite tv


estrecho / madiaq
la plana
northern england
nottingham imc
united kingdom

Latin America
chile sur
cmi brasil
cmi sucre
puerto rico


South Asia

United States
hudson mohawk
kansas city
minneapolis/st. paul
new hampshire
new jersey
new mexico
new orleans
north carolina
north texas
rogue valley
saint louis
san diego
san francisco
san francisco bay area
santa barbara
santa cruz, ca
tampa bay
united states
western mass

West Asia


fbi/legal updates
mailing lists
process & imc docs