🚀 go-pugleaf

The "frame maker" postscript driver does something like this.  What
they do is to encode an image as a single byte of command followed by N
bytes of argument.  The "image" procedure reads the first byte and
looks it up in a table of procedures, then executes the procedure,
which must consume the "right" amount of input (maybe they have a
length field in there too).  The procedures do things like replicate a
given 8-bit pattern, fill in with zeros, fill in with ones, etc..  The
processing time for this is a few percent slower for the interpreter
overhead, but most images spend all their time scaling the image, so it
isn't too signficant.  This kind of compression can help a LOT if an
image has significant regions of constant gray (can you say "margins"
boys and girls?).

Have fun,