Blur Tools for Signal
I don't doubt that Signal has put a significant amount of effort into making sure its blurs can't be reversed. But when I look at the results in the photo, I don't understand why they would put in that effort. Why derive a blur from the base photo at all?
Things that seem way easier to me:
A) blacking out the face entirely with a solid color,
B) if that looks ugly, replacing it with some kind of clip-art,
C) if that still looks ugly, replacing it with a gradient
D) if that still looks ugly, replacing it with a pre-blurred face from a generic set of buckets.
I sort of get the aesthetic argument, but I also really don't, because the way Signal is blurring faces is ugly, at least in the photo they show. It's not a seamless thing that blends into the background and looks way better than a solid color. It's giant squares, and the amount of blurring means that the contents are basically indistinguishable from a radial gradient to my eyes anyway. Am I missing something? Would a gradient really look any worse than this?
Is there some kind of use-case where blurs give aesthetically much better results than what we're seeing in the photo? Are the concerns I'm seeing below about de-masking just fear-mongering? Are blurs in general just pretty safe, fast, and easy to do? Moxie isn't stupid, I assume in situations like this he knows what he's doing.
In case anyone feels like playing around with it, a friend and I made a project to do auto-blurring of faces with OpenCV a few years ago, with both iOS and node frontends ..
iOS module:
https://gitlab.com/seclorum/groupie/-/tree/master/ios/groupi...
Main node.js app:
What kind of blur is used? Blurs are annoyingly bad at obscuring things like faces. They may be good at making faces unrecognisable to people, but they’re not nearly as good at making faces unrecognisable to machines.
Would it be practical to take a facial recognition algorithm and use it to warp the identifying characteristics of faces in a scene such that the faces lose enough uniqueness to make facial recognition ineffective?
My understanding of facial recognition is that it operates on relative positions of facial elements. If you can "delete" this uniqueness from the source material by warping faces towards a limited handful of generic shapes, you make the video less useful to Government intelligence.
You could still blur the result, but you might be able to get away with less blur. Remember that it's important to see that people have faces otherwise they can be more easily dehumanised.
I've seen it said that blurs can relatively easily be reversed. I wouldn't expect that to be unknown to the Signal team, so I wonder if anyone knows how they dealt with that. A different blur method that is not reversible?
Honestly, I can't keep up with acquisitions, full e2e encryption claims, then those claims get debunked, and you can't find out what the truth is.
Based on all information out there, in year 2020, what is the most secure IM app?
What do you recommend to your friends if they care about privacy?
Can someone school me as to why we'd use blur when you can just put a solid block of pixels of the same color over the face? The hard part is face detection, right?
I've tried this feature out and found that it doesn't do as good a job of blurring faces as I'd like, especially when those faces take up more of the frame. I posted some pictures here:
http://lelandbatey.com/projects/signal_blur_comparison/
Basically, I think they're using a constant blur size which fails to adequately obscure faces that take up a lot of the image because when a face takes up a lot of the image then the features of that face become large, which would require even MORE blurring to obscure. And they're not doing "more blurring" when the area which needs blurring grows, or at least they aren't doing enough additional blurring.
They are also distributing physicals masks? It's not even a filtering type mask is it? How odd.
Is the blurring some type of encryption that the user can unblurr or is this a one way road? I am just thinking off some odd circumstance where say they realize they had a picture of a vandal somewhere. But I guess you can then be forced to unblurr everything by law enforcement which might be undesirable in some cases.
Slight off topic from the article, I was reading about the sting ray discussion here on HN yesterday. Signal supports some sort of mesh network communication right? Is that a work around for sting rays? Thanks.
instead of bluring faces we should be replacing them with computer generated faces, double up on fuzzyness and destroying the possibility of easily detecting "its been blurred, i must then take out my best guessing tools then"
Why not just block it out instead of blur? Like all out white or black block, or any color, or a 'redacted' button?
Impressive how quickly they've reacted.
I doubt blurring faces on photos you take helps much in hiding your identity from the government.
In events like these, they likely have access to quite a few image sources that do not blur faces.
So, given an image with blurred-out faces, they can look in those sources for images showing persons with similar skin color, hair, length, and clothing to the person(s) they’re interested in, and from there find your face.
If they are willing to make an effort, even individuals may be able to do that, using photos that people who don’t blur faces upload to the internet.
So an organization that gets millions of dollars from the US government is supporting people that wish to overthrow the government, many of whom are outright anarchists. Seems like a poor investment. Unless....
This is rather silly, you could always draw solid colors over someone's face and it works better than blurring. A rather frivolous update, from a software standpoint. The sentiment is nice.
Is the pattern on these masks meant to confuse facial recognition algorithms or is it just for looks?
I don't know much about image processing, but can't blur from some area in an image be "removed" so as to recover the original image underneath or am I just totally mistaken about how images/pixels work?
As a software engineer: screw software-based solutions. Too hard to communicate to people, too easily compromised without notice, just blegh for things like this.
I remember the Mueller report being printed out, inked over, and then scanned before exported as PDF just to make sure there's no software shenanigans. I really like this idea.
If you wanted to implement that in the field, you could purchase a Polaroid camera, ink over faces manually, and then use your iPhone and take a picture of that picture and destroy the film afterwards.
You can be identified by gait alone.
wouldn't it be easier and more secure to put a noise filled rectangle over the faces?
I don't understand the need for this. There is nothing criminal or embarrassing about being in public or participating in a peaceful protest. Why is this feature needed?
what about videos.
why not use a black square
Cool ! Now stop with the forced contact discovery.
If you decide to block traffic and scream things in public, why should those images be modified? How is that not historical revisionism, like 1984?
Www.linkdin.com
At some point he have to assume this is about defending people that are committing crimes. Nice to see that the radicalisation caused by left wing social media is finally getting to its final conclusion.
Why would peaceful demonstrators need to hide their identity?
I have been to numerous peaceful protests in the US, even been attacked by observers, and have never had to hide my identity.
Additionally, in a large crowd where most will not hide identities, this app is useless.
Only use case I can imagine is a one to many communication likely to be frowned on by authorities, which sounds like the coordination of illegal activity, such as violence and looting.
I wonder if any website where such techniques are popularized would consequently be considered an accessory to whatever illegal activity is being coordinated?
And even if not, as owner of such a platform, it would not rest easy on my conscience to know my site is being used to help coordinate activity that will hurt and harm a great many innocent people.
Without sold proof this is not possible to circumvent, this maybe more dangerous than not.
Here’s an example of AI being able to identify a blurred face: https://twitter.com/ak92501/status/1267609424597835777
Identifying an individual is not just about a face, but number of factors that are much more complex and very hard to account for in a systematic way.
—-
If Signal is really concerned about allowing individuals to control the information they leak, they need to prioritize releasing the feature that will allow users to use Signal without providing phone numbers; one of their staff recently publicly stated this is finally likely to become a feature. Not to mention stop repeatedly asking for the user to provide their name, access to contacts lists, etc.