Efficient view synthesis for multi-view video plus depth
In 3D video, depth-image-based rendering (DIBR) is widely employed in view synthesis to generate virtual views. However, the processing of this algorithm is based on the frame-level, and the characteristics in different regions cannot be fully taken into account before rendering. This drawback will lead to the unnecessary and redundant information in some regions being abused, which increases extra computation. This paper proposes a region-level 3D-warping method for DIBR, where regions are divided according to their characteristics. Then, only the necessary information in some important regions is utilized warping so that the redundant information could be avoided in the computation. Experimental results show that our approach is almost 4 times faster than VSRS-1D-fast, while declines 0.12 dB PSNR in the performance of synthesis views averagely. Hence, our method can achieve a good trade-off between the computation and view synthesis and will be especially useful for applications where the computation is the concern.