Had to manually download my entire library (5 items at a time with no useful labels) before requesting account deletion. I’m glad they’re making account deletion easy, but disappointed that it had to go.
Had to manually download my entire library (5 items at a time with no useful labels) before requesting account deletion. I’m glad they’re making account deletion easy, but disappointed that it had to go.
I used a large-tile mosaic blur then motion blur. I’d be very impressed if someone managed to recover my account info from a destructive process like that.
There’s algorithms for that. Yes that’s blur from actual motion but a lot of the math carries over. Blur is not considered destructive because all the information is basically still there. Not a big deal for the email in this post probably, but anything more sensitive? Yeah, nah, blur isn’t enough
…do you know what a mosaic blur is?
Do you? Actually there’s other algorithms I’d try first if I know the font, which I’d just look at Humble’s CSS to see it’s default font first.
People are just trying to make sure in the future you don’t accidentally give away $1000 and you’re being an ass about it. (for the purposes of undoing your blur we could use the @ symbol to tell if we’re on the right path like they did with the known characters of the private key).
Thank you, very interesting! I never thought large tile sizes / blur radius would be recoverable :O
Thanks for sharing all this info!
Of course! This sort of image processing is what I wanted to do when I originally went for my computer science degree so I’m always glad to talk about it. Life took turns so I went a much different direction for work but it’s still something I enjoy.
A simple pixelization and a mosaic are not the same thing in raster image editing. You’re barking up the wrong tree. The algorithms you linked can’t recover based on what’s left in the image, since over half of it is missing and the other half isn’t a simple pixel grid. The mosaic distort introduces additional noise that prevents being reversed when the appropriate parameters are applied.
It’s a known character set and font in a known order and alignment. We don’t give a shit that half the pixels are missing, we could be down to 5% of each character and still recover it. What’s left will still blur differently based on what the individual characters are, all the way down to giving us testable hints based on character width.
The first paper I linked is literally “On the ineffectiveness of mosaicing”. You’ll have to explain how you’re defining these things different from the author when they’re using pretty standard terms and discussing image editing techniques used on social media to redact personal information from images, literally the thing you did.
You’re making claims, being disproven, and have nothing more than “I know more than you” as a response here.
The first paper you linked is about simple pixel density decrease, which is not the same thing as a mosaic distort. The term “mosaic” referring to simple pixelization is an outdated term in the image editing world. You can use mosaic tools to pixelize (by limiting it to squares only and normalizing the effects layer) but that simply isn’t what the term means anymore.
Just for fun I found the code gimp uses for that
I’m just on my phone so I’ll admit I might have overlooked something in there but it looks like they draw a grid (be it triangles or squares or voronoi), then gaussian blur (also called a convolution) inside the cells to average the color. This is exactly the sort of stuff I’ve been talking about and basically exactly what I expected their code to look like.
Is it possible you distorted enough to be unrecoverable? Maybe, but I know I wouldn’t bet my money/info on it.
That mosaic distort is still just pixel averaging with voronoi noise. The basic principle is the same. Deconvolution and go.
Do you know what a chosen-plaintext attack is?
So no, you don’t. Got it.
Edit: You think you can reverse a destructive editing technique that deliberately creates gaps in the original, leaving too little information to reverse, I invite you to try. I can tell you right now that you won’t get far. You can pretend you understand these techniques all you want, but as long as you maintain that a process that deletes over 90% the original information can be recovered in full you’re just showing off how stupid you are.
I think I can run strings through the same visual hash until I find a match, but you do you.
Based on just your comments here, you have no right to judge anyone else’s intelligence.
Emails have two known symbols in them, the ‘@’ and the ‘.’, and you’ve said what two blurs you used. I would have called it a birthday attack myself (where the blur steps are the encryption key and depends on letters roughly blurring differently, aka hashing differently).
Of the people here they’re not the one showing what they don’t know
Aaaaaaaaackthually… You do not need a dot. Putting an MX record on a TLD is perfectly valid. Though pretty much 100% of all email validation forms are doing it wrong and requiring it. So in this case it almost certainly has a dot.
Oh what’s technically valid as an email is insane. I’ve looked into it before for work and went “nope” and just grabbed one of the standard “wrong” regexes everyone uses.
Yepp, the spec is super weird and too complex. Like @ is allowed if quouted in the local part.
Almost no one has a MX on TLD though so most just reject it, you can also argue that it is a good thing to check for. If someone writes “gmail”, they really just forgot the .com part so having the UI warn about it makes sense.
The best way to validate is to send an email to it.
You’re saying that this is reversible when you don’t know any of the parameters?
Have fuckin fun proving that.
With text? Yes absolutely. The font is all I need, I will calculate the parameters based on how I observe the text to have been modified.
Bet, post your credit card info and I’ll put in the effort
They don’t need to recover the full image, and those filters aren’t really destructive, especially the blur/smudging. There’s maybe 60 likely characters in an email address. The font is known and common mosaic and blur algorithms are known. You know there’s in @ in there somewhere and likely ends with a .com or other common tld. You can do things like reverse the blur, then create a lookup table of mosaics to alphanumeric characters and decode the text that way.
It’s more complicated than that, and nobody’s gonna bother to decode this image, but it is 100% possible and has been done before. Using a solid color is easier, faster, and more effective.
Do you know how a mosaic blur works? There is a 0% chance anyone is decoding more than the 5 characters of @, ., c, o, and m.
Do you? https://codeberg.org/spipm/Depixelization_poc/
I don’t! Got a link to the algorithm?
You might know it better by it’s other name of pixelisation
A very basic algorithm would be:
Not quite.
That’s not a mosaic blur which is what you said you used. So, just kind of proving our point you don’t know what you’re talking about, also that’s just as reversible for a mosaic blur on text as the pixel averaging is the important part and still done
Photoshop and Affinity (and Photopea, iirc) place the equivalent filter under Filters > Blur > Mosaic. But you’re right, it’s not a mosaic blur. Despite being named “mosaic”. Under the “blur” filter category. Not at all.
Huh? Why would you shut down your frontal lobe and go into a spiral of defensive maneuvers over someone helping you with security?
Just by eye I think I can make it out. The domain looks like [check dms] to me.