Distributed video coding with decoder-driven skip

In Distributed Video Coding (DVC), compression is achieved by exploiting correlation between frames at the decoder, instead of at the encoder. More specifically, the decoder uses already decoded frames to generate side information Y for each Wyner-Ziv frame X, and corrects errors in Y using error correcting information received from the encoder. Typically, this error correcting information is requested using a feedback mechanism in which the decoder requests portions of bits from the encoder until the frame is reliably decoded. In this paper we propose new techniques for deciding when a frame is reliably decoded, based on estimated decoded quality. We define criteria both for the coefficient level and the bitplane level that will allow us to skip coefficient bands and bitplanes. If bitplane decoding is nonetheless needed, the bitplane-level criterion allows early termination of the decoding process, by focusing only on a subset of the bits instead of all. Results indicate significant improvements especially for sequences that can be well predicted at the decoder side.