Keeping only the unique rows in a cellular automaton (CA) reduces redundant information. This works best for class 1 and 2 CAs, those that involve only one color and repetitive patterns, respectively. The CA classes 3 and 4 (those that behave randomly and with local random patterns, respectively) cannot be compressed in most cases.
To find redundancy in CA patterns that evolve laterally or with inclination, it is useful to shift or transpose the data.
The process of compressing a CA is not reversible.