Correlation Noise Estimation in Distributed Video Coding
Jrgen Slowack, Jozef korupa, Stefaan Mys, Nikos Deligiannis, Peter Lambert, Adrian Munteanu, Rik Van de Walle · InTech eBooks · 2011
Video compression is achieved by exploiting spatial and temporal redundancies in the framesequence. In typical systems, the encoder is made responsible for exploiting the redundanciesby predicting the current frame to be coded from previously coded information (such asother frames and/or blocks). Next, the residual between the frame to be coded and itsprediction is transformed, quantized, and entropy coded. As the quality of the prediction hasa large influence on the coding performance, high performing but computationally expensivealgorithms for generating the prediction have been developed. As a result, typical videocoding architectures show an imbalance, with an encoder that is significantly more complexthan the decoder.A new way for performing video coding has been introduced in the last decade. This newparadigm, called distributed video coding (DVC), shifts the complexity from the encoder tothe decoder. Such a setup facilitates a different range of applications where the main focusand constraints are on the video (capturing and) coding devices, instead of on the decoding(and displaying) devices. Some examples of target applications include video conferencingwith mobile devices, wireless sensor networks and multi-view video entertainment.The aforementioned shift in complexity is realized by making the decoder responsible forgenerating the prediction, hereby relieving the encoder from this complex task. While theencoder has the ability to select the best prediction based on a comparison with the originalto be coded, the decoder can not perform this comparison as it has only access to alreadydecoded information, and not to the original. This complicates the decoder’s task to estimatean accurate motion field compared to conventional predictive video coding.In distributed video coding, the prediction generated at the decoder (called the sideinformation) often contains a significant amount of errors in comparison to the original videoframes. Therefore, the errors are corrected using error correcting information sent by theencoder (such as turbo or LDPC codes). For efficient use of these error correcting bits, softchannel information is needed at the decoder concerning the quality of the generated side