Hacker News new | ask | show | jobs
by apodolny 1838 days ago
Yes - this is definitely an issue, especially with text data. One thing I’m curious about: Google Street View apparently adds noise and then blurs. Is this a viable option?
2 comments

If you really really want blur, just apply enough of it. Otherwise, just pick a solid color that matches the environment.

The text example in the article that GP linked[1] looks pretty reversible to me indeed. Not sure it needs a neural network, or at least it could be enhanced a lot with character frequency checking or matching words against a dictionary, but I haven't ever seen text unblurred where I didn't expect it might be possible.

Personally I don't find blurring to be less annoying than a reasonable color. Pitch black stands out a lot, but something close to the background color (but clearly distinct) is unobtrusive while also being clear that something was censored and not just a broken image.

[1] https://hsto.org/storage2/eff/36d/77a/eff36d77a583b46e461c12...

When dealing with text, couldn't the software replace the area with some text block of Lorem ipsum-style gibberish before blurring it? It could even try to be a little clever and mimic the format of the original text. This way you get the desired effect with absolute secrecy, with the added bonus that you bait people who try to unblur it into wasting their time. :)
Certainly the text not being under the blur would be the best way to hide what was under the blur. The trick is of course in actually making it do that in an automated fashion.
If I wanted to keep the blur aesthetic, I'd probably do a full removal, run an inpainting algorithm to replace the removed region with something less jarring in context and then blur the result. The inpainting algorithm can be fairly low quality and still get acceptable results since it won't be seen directly.