Fusion side information based on feature and motion extraction for distributed multiview video coding

In distributed multiview video coding (DMVC), the quality of side information (SI) is crucial for decoding and the reconstruction of the Wyner-Ziv (WZ) frames. Generally, its quality is influenced by two main reasons. One reason is that the moving object of the WZ frames can be easily misestimated because of fast motion. The other is that the background around the moving object is also easily misestimated because of occlusion. According to these reasons, a novel SI fusion method is proposed which exploits different schemes to reconstruct different parts complementarity. Motion detection is performed to extract the moving object which can be predicted by utilizing both temporary correlations and spatial correlations. As for background around the moving object, temporary correlations are utilized to predict it. It is noteworthy that the prediction method used in this paper is based on a feature based global motion model. The experiment results show high precision quality of the SI of the WZ frames and significant improvement in rate distortion (RD) performance especially for the sequence with fast moving objects.