Extractable perfectly one-way functions

We propose a new cryptographic primitive, called extractable perfectly one-way (EPOW) functions. Like perfectly one-way (POW) functions, EPOW functions are probabilistic functions that reveal no information about their input, other than the ability to verify guesses. In addition, an EPOW function, f, guarantees that any party that manages to compute a value in the range of f "knows" a corresponding preimage. We capture "knowledge of preimage" by way of algorithmic extraction. We formulate two main variants of extractability, namely non-interactive and interactive. The noninteractive variant (i.e., the variant that requires non-interactive extraction) can be regarded as a generalization from specific knowledge assumptions to a notion that is formulated in general computational terms. Indeed, we show how to realize it under several different assumptions. The interactive- extraction variant can be realized from certain POW functions. We demonstrate the usefulness of the new primitive in two quite different settings. First, we show how EPOW functions can be used to capture, in the standard model, the "knowledge of queries" property that is so useful in the Random Oracle (RO) model. Specifically, we show how to convert a class of CCA2-secure encryption schemes in the RO model to concrete ones by simply replacing the Random Oracle with an EPOW function, without much change in the logic of the original proof. Second, we show how EPOW functions can be used to construct 3-round ZK arguments of knowledge and membership, using weaker knowledge assumptions than the corresponding results due to Hada and Tanaka (Crypto 1998) and Lepinski (M.S. Thesis, 2004). This also opens the door for constructing 3-round ZK arguments based on other assumptions.