View synthesis based on temporal prediction via warped motion vector fields

View synthesis method for 3D video coding based on temporal and inter view correlation

“…In fact, there is also the temporal correlation between frames inside the synthesised view which can significantly improve the quality of the synthesised view. For instance, in [11][12][13][14], the authors compute the motion vector fields (MVFs) of the reference views and warp vectors into the synthesised view. The warped motion vectors are then used to exploit the temporal correlation between frames.…”

“…More specifically, Purica et al [11] and Chen et al [12] use warped MVFs to retrieve information about disoccluded regions from other frames. In [13], the motion compensation (MC) is performed with sub-pixel precision using warped MVFs to obtain temporal predictions which are blended together with the DIBR. The view synthesis method proposed in [14] also use warped MVFs to create temporal predictions for the intermediate view, but an additional frame per group of pictures (GOP) in the intermediate view is required to compress and send as the reference frame for the MC.…”

“…However, because of high precise MVFs requirement, these methods need to use the optical flow or block-based motion estimation to deliver MVFs which may not be appropriate for many emerging applications constrained by the computation resources. For example, in [11,13,14], the optical flow algorithm in [16] is used to deliver MVFs.…”