Stitching of H.264 video streams for continuous presence multipoint videoconferencing

摘要

This paper proposes a video stitching method applicable to continuous presence multipoint videoconferencing. The proposed approach is universally applicable yet computationally simpler than a pixel-domain approach which is computationally expensive due to the need for full re-encoding. A compressed-domain approach is far simpler but its applicability is limited to H.261. The proposed method employs a blend of compressed-domain and pixel-domain tools to eliminate the drift that results when a compressed-domain approach is attempted for H.263 and H.264. Coding parameters from incoming streams are reused on the outgoing side, thereby avoiding the computational bottleneck of re-estimating them. Essential details for implementing the proposed method are presented for H.264, along with simulation results for validating it. Application of the drift-free stitching method to H.263 is provided as an Appendix A.