Skip to Main Content
The distributed source coding theory teaches that efficient compression can be achieved by exploiting the correlation between encoded sources at the decoder only. A critical issue in this coding paradigm is how to accurately estimate the correlation statistics at the decoder. This paper proposes a novel correlation channel estimation method designed for generic layered Wyner-Ziv (WZ) coding. Driven by real-life applications, i.e., video coding, the correlation noise is assumed to be memoryless zero-mean Laplacian with a block-stationary behavior. Unlike existing methods, the proposed technique derives a novel maximum likelihood estimate of the noise scaling parameter per stationarity block. Adhering to layered coding, the derived estimate is successively refined as more stages are decoded. The Crámer-Rao lower bound (CRLB) for our estimator is derived and the mean squared error performance of the proposed successive refinement algorithm is evaluated through simulations. Apart from achieving an accuracy close to the derived CRLB, the proposed algorithm is shown to yield a WZ rate-distortion performance approaching the ideal but unrealistic scenario where the decoder knows a priori the correlation statistics. Moreover, the proposed technique has been integrated into our latest WZ video codec for wireless capsule endoscopy. Experimentation reveals that the proposed method delivers improved compression performance compared to state-of-the-art techniques, while simultaneously reducing the overall decoding complexity.