zlacker

> How resilient is blurring against deconvolution?

This depends a lot on the implementation details. If you blur an image using arbitrary-precision real numbers, then blurring is invertible. If you add a bit of random noise, or quantize your pixels into a finite-precision data type, then it becomes essentially one-way, and you cannot recover the original image.

>>enriqu+(OP)
> you cannot recover the original image

Technically you are correct - you cannot recover the exact original image. The same is the true for saving an image as JPEG. But the question at hand is whether you can still recognize faces, not whether you can restore a byte-for-byte of the original. And whether JPEG or blurring, the answer is generally "yes".

It does depend on the implementation (and whether you know the implementation) how close you can get.

>>Mauran+jg
Yes, my point is that it depends a lot on the specific blurring. If you just average a square neighborhood of 4 pixels around the center, and add no noise, it is very likely that you can "enhance" the resolution back to almost the original image. Yet, if the blur kernel is much larger (say, a gaussian of width 40 pixels), and you add some noise after te blur, it is very likely that you have completely destroyed the information.