Distributed keyless security for correlated data with applications in visual sensor networks

Abstract

This paper examines the problem in which several nodes sharing highly correlated data, such as visual data, wish to compress and encrypt their data to provide confidentiality. The nodes however perform these tasks separately, without communicating with one another and without the use of cryptographic keys. The base station (BS) receiving all such encrypted data, can reconstruct each of the nodes' data, whereas a passive eavesdropper who is only allowed a subset of the encrypted data gleans as little information as possible about the nodes' data.We build on previous results with the goal of increasing secrecy (measured by Shannon equivocation) by: (1) relaxing the BS's perfect reconstruction criterion thus permitting non-zero distortion reconstruction; (2) permitting communication (feedback) from the BS to the nodes. We show that permitting non-zero distortion reconstruction does increase secrecy, however unconditional secrecy is still not achievable unless the distortion is maximal. We also prove that feedback from the BS usually (under most practical scenarios) does not improve secrecy, unless the BS has certain knowledge concerning the eavesdropper.Finally this paper proposes ideas for applying the results to images by analyzing the ideal image model to demonstrate the practical difficulties in achieving provable security for images.