[kundor]

Yes, to an extent that would work.

1) Take target image

2) Find the index in π (say) where a matchable sequence of bytes occur, like for Waldo. (If your image is big, this could take several years of computation time.)

3) Transmit the palette, width, height, and index: about 800 bytes.

4) Probably the index is too far out for everyone to have a copy of the data already (it would be beyond terabytes). So the recipient then spends several years computing the number out far enough.

5) Profit!

Note: the palette is optional; you could leave it out and only transmit about 32 bytes. Using a palette also means lossiness, because you're reducing to 256 colors and making compromises between pixels. But using one saves you several orders of magnitude of computation time.

[kundor]

PS. The fact that any image can be encoded in 32 bytes this way implies that there are only 2^256 possible images — about one hundred and sixteen quattuorvigintillion. That's obviously not literally true, but we are searching for a subsequence of π which is "close enough"; the number of possible images which humans would consider distinct is probably well less than that.

(That is, the vast majority of possible images are "color noise" which all look more or less the same.)