Reducing complexity assumptions for statistically-hiding commitment

  • Authors:
  • Iftach Haitner;Omer Horvitz;Jonathan Katz;Chiu-Yuen Koo;Ruggero Morselli;Ronen Shaltiel

  • Affiliations:
  • Department of Computer Science, Weizmann Institute of Science;Department of Computer Science, University of Maryland;Department of Computer Science, University of Maryland;Department of Computer Science, University of Maryland;Department of Computer Science, University of Maryland;Department of Computer Science, University of Haifa

  • Venue:
  • EUROCRYPT'05 Proceedings of the 24th annual international conference on Theory and Applications of Cryptographic Techniques
  • Year:
  • 2005

Quantified Score

Hi-index 0.00

Visualization

Abstract

Determining the minimal assumptions needed to construct various cryptographic building blocks has been a focal point of research in theoretical cryptography. Here, we revisit the following question: what are the minimal assumptions needed to construct statistically-hiding commitment schemes? Previously, it was known how to construct such schemes based on one-way permutations. We improve upon this by constructing statistically-hiding commitment schemes based on approximable-preimage-size one-way functions. These are one-way functions for which there is an efficient way to approximate the number of preimages of a given output. A special case (for which we show a somewhat simpler construction) is that of regular one-way functions where all outputs have the same number of preimages. We utilize two different approaches in constructing statistically-hiding commitment schemes. Our first approach proceeds by showing that the scheme of Naor et al. can be implemented using any one-way function having an output distribution which is “sufficiently similar” to uniform. We then construct one-way functions with this property from approximable-preimage-size one-way functions. Our second approach begins by constructing a commitment scheme which is statistically hiding against an honest-but-curious receiver. We then demonstrate a compiler which transforms any such commitment scheme into one which is statistically hiding even against a malicious receiver. This compiler and its analysis may be of independent interest.