Practical Distributed Video Coding based on source rate estimation

In Distributed Video Coding (DVC) the encoder compresses frames at source rates that depend on the statistical dependency between the Wyner-Ziv (WZ) and side information frames. An important issue that we address in this paper is providing the encoder with a mechanism to identify the source rate to be used in encoding a WZ frame. One possible solution is to follow a feedback approach; the encoder starts by sending a small amount of data (low source rate). In case the decoder fails to recover the compressed frame, it provides the encoder with a feedback requesting more bits. This solution results in using source rates that are ideal in the sense that they are the minimal rates that lead to successful decoding. Nevertheless, this solution might not be practical for visual sensor networks as it may exhaust limited bandwidth and energy resources. We propose a mechanism to estimate ideal source rates without the need to exchanging feedback messages. The proposed mechanism uses conditional entropy (entropy of a WZ source conditioned on a side information source) to estimate ideal source rates. Further, we show that by estimating the source rates (using optimal estimators) we can achieve a video quality and compression performance that is close to that achieved when feedback messages are exchanged. Moreover, we show that by avoiding incremental transmission and feedback messages, the proposed estimation-based approach can demonstrate lower energy consumption for a range of video quality compared with the feedback approach when both are deployed over a real visual sensor platform, namely, the imote2/IMB400.