Depth-image-based rendering (DIBR) is a key step in 3D video generation. Parallel implementation of DIBR is able to improve rendering efficiency. General DIBR algorithms include two steps: pixel shifting (warping) and hole filling. There are memory correlations in these steps. To minimize memory conflict, we employ an auxiliary matrix to record maximum shifting distance. Implementation details on OpenMP and CUDA are presented and experimental results on GPU and multi-core CPU are compared.
展开▼