Dheera Venkatraman

https://dheera.net -- mostly photography and random non-work side projects, sorely in need of update

This is a great idea! You might reconsider using blur to obscure the information, as it is trivial to defeat [1] (no ML required!). [1]: http://dheera.net/projects/blur

Not that I'll be doing this, but... The blurred areas on the ROV video (https://www.youtube.com/watch?v=3p8TH1tDvzk ) probably contain location data, among other things. Considering that we have a nice, nearly black background, and access to the font used, it should be a piece of cake to figure out the censored data. Just blur numbers 0-9 and determine which fit best with the video. See e.g. http://dheera.net/projects/blur

and not by just drawing that box over them in the original (textual content) pdf. You need to squash it down to images, then destroy the bits you need to. (There might be other ways, but enough Big Government Agencies have screwed it up to make it worth noting) Is the 'decoding' paper this one: http://dheera.net/projects/blur.php ? The blur function is just smearing pixel values across their neighbours in blocks, so you can treat it as a hashing function, and then generate enough candidates that eventually you get something that hashes correct (or close enough) I'm not sure how practical it'd be on data longer than a credit-card number, but it's an interesting hack nevertheless.

Tip: blurring photos of sensitive information is a bad idea: https://dheera.net/posts/20140725-why-you-should-never-use-pixelation/ https://news.ycombinator.com/item?id=8078747

I suspect there's not much information in the individual blurred face, but I wonder if given enough examples you'd be able to determine if an unblurred face is the one in a sample of images with any level of confidence? You can do that with text (http://dheera.net/projects/blur).

Just a heads-up - the pixelation used to hide the keys in the final screenshot is not safe! There aren't that many possible characters in each of those 10x10px blocks, and ~256 levels of average brightness is enough to distinguish between them. (Here's a simplistic demo: https://dheera.net/projects/blur) Stick to the block-of-constant-colour method used in the first few screenshots. It's ugly but safer!

Check out Deconvolution[1]) The technique has been done before and the text string of the private key would probably be the easiest point of attack. WHY BLURRING SENSITIVE INFORMATION IS A BAD IDEA[2] RESTORATION OF DEFOCUSED AND BLURRED IMAGES[3] Take away: black out any and all sensitive information. Blurring just marginally reduces the independence of the bits presented in blurred output. To have privacy, all dependence must be eliminated, e.g. black out. [1] https://en.wikipedia.org/wiki/Deconvolution [2] https://dheera.net/projects/blur [3] http://yuzhikov.com/articles/BlurredImagesRestoration1.htm

Just another [0] example of why you shouldn't blur sensitive info but completely black it out. [0]: https://dheera.net/projects/blur

Pixelating and blurring images has long been known to be insufficient in completely obscuring information [1]. In fact, in a lot of computer vision work, the image resolution is reduced to ignore noise and facilitate the workload for algorithms anyway. Completely destroying information through blacking out is preferable. [1] https://dheera.net/projects/blur

Blurring or mosaicing is not very secure at all [1]. To be more safe block out sensitive information with solid black. (Just remember to not leak any information in the metadata, like an embedded thumbnail.) [1]: https://dheera.net/projects/blur

While stego might be a route, blurring the watermark may not be enough to obscure the source. Given a known frame, they could blur the same portion of the same frame on all watermarked copies, and the one with a matching output is your culprit. Blurring loses information, but is more akin to a hash than a deletion, as it's a deterministic process. Edit: this sort of thing. http://dheera.net/projects/blur

Maybe something of this sort? http://dheera.net/projects/blur Just get the font and do comparison on what fits, if there's too many results, you could use a dictionary.

Please, don't blur if you want to redact. Instead, use a uniform, opaque color. See http://dheera.net/projects/blur and https://news.ycombinator.com/item?id=8078747. Context: I just looked at some of the screenshots showing example findings. While it is thoughtful to blur some sensitive information, it is clear that blurring is not enough. I hope that we can get this message out.

For the record, blurring (even when applied appropriately) is still a pretty bad idea for hiding information [1]. I know this has been on HN a few times. I'm not sure what you can do with the GA key or if it's even private, but just adjusting levels in gimp shows the numbers. [1] http://dheera.net/projects/blur

Now, does anyone have the time to figure out how many users Github has in their users table? :) http://dheera.net/projects/blur

That level of blurring is almost certainly insufficient to actually redact the information, as discussed in http://dheera.net/projects/blur.php (discussed here http://news.ycombinator.com/item?id=1939607)

See also https://dheera.net/projects/blur

Good questions. My thoughts: - Are you asking about recovery of information that was intentionally blurred? If so, there have been some exciting results in this area in recent years. For example, it turns out that given reasonable priors on the underlying glyphs, it's possible to reconstruct even heavily blurred text. (e.g., http://dheera.net/projects/blur.php ) That doesn't involve deblurring per se, since the goal is simply to find the glyphs (and combinations thereof) with maximum likelihood given the blurred text. Still, it's interesting, and similar techniques have been applied to super-resolution problems lately, where the approach is often called "hallucination." - Blind deconvolutions often are used by experts. However, it's possible to help things along if a glint exhibiting the blur of interest is visible in the photo. That is: if a bright dot of light is visible in the photo, it will be blurred too, and so in the blurred photo the point of light will look like the point spread function (the blur kernel), and that can then be extracted and used with a non-blind deconvolution approach to recover the latent sharp image. - Blind deconvolution and super-resolution are related problems, so advances in one might certainly help in the other. The research you cited makes use of image hallucination (mentioned above) as a key part of its super-resolution algorithm.