Motion Compensated Interpolation Under Complex Motion And Its Application In Wyner-Ziv Video Coding

In traditional video coding standards such as MPEG or H.26x, correlation between sequence frames in the coding side is used to carry out inter-frame predictive coding. In order to predict interpolated frame accurately, motion estimation algorithm is used in the existing standards. Since the complexity of motion estimation algorithm, the complexity of video encoder is 5 to 10 times higher than the decoder. This character is suitable for streaming video-on-demand systems and broadcasting, because that the task of terminal in these applications is low-complexity decoding. In some special application systems, such as wireless video sensor networks or satellite video communication systems, which transmit information of video to the server by low-complexity encoder, but the traditional video coding schemes are not suit for these applications. Therefore, distributed video coding is proposed to resolve the above problems. Some sources with correlation use the encoder which is independent to each other to encode the bit-stream and then send to a single decoder, which operates jointly on all incoming bit-streams in distributed video coding and takes motion estimation at the decoder to decrease the complexity of encoder.In DVC system, one of the key issues is how to construct accurate side information. The accuracy of side information would essentially influence rate distortion and compression efficiency of the DVC system. Most of motion estimation algorithms are based on the assumption that trajectories of objects are translational in a period of time, but are not always linear, which leads to the result that side information is not accurate enough. Hence, the study of nonlinear motion trajectory to obtain the accurate side information has practical significance.With a detailed analysis of the algorithms of side information estimation, this paper proposes an algorithm of motion compensated interpolation with acceleration, which uses a quadratic motion trajectory model that incorporates both velocity and acceleration. The algorithm for the estimation of velocity and acceleration fields in the formulated problem uses Gibbs-Markov models that are linked together by the Maximum A Posteriori (MAP) probability criterion. To decrease the complexity of the algorithm, a multi-resolution approach is used. Experimental results show that the proposed method performs better than the existing methods in PSNR and perceptual quality.