An Analysis of the Blockcipher-Based Hash Functions from PGV

Preneel, Govaerts, and Vandewalle (1993) considered the 64 most basic ways to construct a hash function H: {0, 1}*->{0, 1}(n) from a blockcipher E: {0, 1}(n) x {0, 1}(n)->{0,1}(n). They regarded 12 of these 64 schemes as secure, though no proofs or formal claims were given. Here we provide a proof-based treatment of the PGV schemes. We show that, in the ideal-cipher model, the 12 schemes considered secure by PGV really are secure: we give tight upper and lower bounds on their collision resistance. Furthermore, by stepping outside of the Merkle-Damgard approach to analysis, we show that an additional 8 of the PGV schemes are just as collision resistant (up to a constant). Nonetheless, we are able to differentiate among the 20 collision-resistant schemes by considering their preimage resistance: only the 12 initial schemes enjoy optimal preimage resistance. Our work demonstrates that proving ideal-cipher-model bounds is a feasible and useful step for understanding the security of blockcipher-based hash-function constructions.