Comment by bastawhiz

There is far less information in the artifacts than in the part of the image that was redacted. And of course, you need to replicate the process of redacting the original image, which requires you to know about the algorithm that encoded the image. It also, if I'm not misunderstanding, requires you to have a version of the redacted image without the artifacts from the redacted section (or a second identical version with a different redacted graphic, like the red and blue circles)

Do you have a proof of concept of the technique you're describing? Otherwise I remain skeptical.