Correlation channel estimation in pixel-domain distributed video coding

The paper addresses an essential problem in distributed video coding (DVC) which is modeling and estimation of the correlation channel. Current works assume the distribution of the correlation noise to be independent from the realization of the side-information. The paper demonstrates that this assumption is inaccurate and proposes a novel model which depends on the realization of the side-information. The proposed model is experimentally validated showing remarkable accuracy over conventional models. Driven by the side-information-dependency of the proposed model, a novel technique for estimating the correlation channel in video at the decoder is introduced. The proposed technique is incorporated into a unidirectional spatial-domain DVC system achieving similar performance compared to the state-of-the-art transform-domain DVC at a much lower encoding complexity.