A VIDEO OBJECT SEGMENTATION SCHEME TO LOWER COMPLEXITY IN MULTI-VIEW VIDEO CODING

摘要

Exploiting redundancy is the hallmark of traditional video coding but is even more essential in multi-view video coding (MVC). The exploitation of additional redundancies, however, incurs extra computational complexity, counteracting the benefits gained from coding efficiency. In this paper, we propose an efficient segmentation scheme for MVC, which aims to lower complexity while achieving high coding efficiency in MVC. The proposed scheme is applied to the base view in MVC using the intensity of decoded frames and motion vectors of base view, which can be also recovered by the decoder. Therefore, no segmentation masks need to be encoded. So that increase of complexity and output bit rate from segmentation can be limited, allowing fast and on-line processing. The segmented objects in base view are then applied to an object-based MVC, which using object registration and warping to predict enhancement views (EVs) and initial disparity maps of EVs. These features are utilized in lowering the complexity of Evs' coding. The experimental results show that the proposed technique provides high quality and consistent segmentation masks with low computational complexity. In addition, by applying the proposed scheme, high coding efficiency and low complexity object-based MVC is achieved.