Compressed domain stitching of HEVC streams for video conferencing applications

We present a method for compressed domain stitching of streams coded according to the (draft) High Efficiency Video Coding (HEVC) specification for video conferencing applications. The main challenges for mixing of video streams in the compressed domain are elaborated and divided into pixel processing level, the level of syntax elements, and entropy coding level. A special focus is on the mechanisms of intra- and inter-prediction mismatch and the respective countermeasures. We propose a solution for mixing video streams on tile level enabling light-weight video stitching. This includes a novel approach for avoiding mismatch due to temporal prediction. The simulation results show a significant gain in compression efficiency and computational complexity over pixel domain stitching of video content.