> For example, if you have a relatively wimpy camera that streams out a 160×120 grayscale image, it might be fun to see how far you can push something like combining:
I'm typically not one for self-promo but I did this with video and you can get 90% compression with exactly those three basic steps. RGB -> YUV420, delta encoding, then basically-upgraded-RLE. You can find the guide here: https://github.com/kevmo314/codec-from-scratch
Yep adding noise would mess up the deltas which would reduce the efficiency. This can be mitigated with a quantizer, which I was hoping to write about next.
I suspect it's possible to achieve 99% lossy compression without getting too complex to be understandable in the toy codec with a couple more of these similar techniques.
I'm typically not one for self-promo but I did this with video and you can get 90% compression with exactly those three basic steps. RGB -> YUV420, delta encoding, then basically-upgraded-RLE. You can find the guide here: https://github.com/kevmo314/codec-from-scratch