Discriminatory Lossy Source Coding: Side Information Privacy

The Heegard-Berger problem models a case in which encoding at a source has to account for two decoders, one with and one without correlated side information when the same information is not available at the encoder. The Heegard-Berger encoding scheme is proved to be rate-optimal even when an additional constraint on the privacy of side information is imposed at the uninformed decoder. The results are illustrated for a binary source with erasure side information and Hamming distortion, a result which is also of independent interest.