How do you compress random data? My idea is that you don’t compress. You can recognize parts of the image that is looking random and replace it with similar random data but from a random number generator. This will however hopefully reduce the storage size.
It will be a lossy compression. But since its still looks like random you will not be able to tell the difference.
Maybe one can call it random image blindness. That is. If you percive an equal quality between a sampled part and the generated part you are blind to the difference.