Or may a zip-style algorithmn, but as the "quality" setting gets lower, unique/low frequency sequences are replaced by the closest common sequence.