Distributed Video Coding with Geometric Transforms

Abstract The Distributed Video Coding (DVC) paradigm is based on two well-known information theory results: the SlepianWolf and the Wyner-Ziv theorems. In a DVC codec, the correlation between video signals is exploited at the decoder, providing a flexible distribution of the computational complexity between the encoder and the decoder and error robustness to the channel errors. To exploit the correlation between frames in a DVC codec, a translational motion model is typically used. However, this model is not accurate enough when complex motion occurs, such as rotations and zooms. In this paper it is proposed a geometric transform based motion model to generate the side information in a distributed video codec. The side information is an estimation of the original frame to code created at the decoder. The proposed Unidirectional Warping Side information (UWSI) creation solution uses a novel motion estimation with geometric transforms. Experimental results show PSNR gains of up to 0.9 dB in SI quality and 0.6 dB in RD performance for some video sequences. The Distributed Video Coding (DVC) paradigm is based on two well-known information theory results: the SlepianWolf and the Wyner-Ziv theorems. In a DVC codec, the correlation between video signals is exploited at the decoder, providing a flexible distribution of the computational complexity between the encoder and the decoder and error robustness to the channel errors. To exploit the correlation between frames in a DVC codec, a translational motion model is typically used. However, this model is not accurate enough when complex motion occurs, such as rotations and zooms. In this paper it is proposed a geometric transform based motion model to generate the side information in a distributed video codec. The side information is an estimation of the original frame to code created at the decoder. The proposed Unidirectional Warping Side information (UWSI) creation solution uses a novel motion estimation with geometric transforms. Experimental results show PSNR gains of up to 0.9 dB in SI quality and 0.6 dB in RD performance for some video sequences.