In this thesis, efficient video delivery solutions are designed to address screen content coding (SCC) and 360-degree video streaming. Towards this goal, multiple sub-problems and applications are addressed.

The first problem is to accelerate screen content encoding and transcoding, while simultaneously preserving the coding efficiency. For encoding acceleration, machine learning based fast algorithms are proposed, using High Efficiency Video Coding (HEVC) coding unit (CU) features for fast block mode and partition decisions. For transcoding acceleration, additional side information from the decoded bitstream is further incorporated for fast block mode and partition decisions, using machine learning and mode mapping techniques for forward HEVC to HEVC-SCC transcoding and backward HEVC-SCC to HEVC transcoding, respectively. Significant encoding and transcoding complexity reductions have been achieved with negligible losses in coding efficiency.

The second problem is to efficiently stream on-demand 360-degree videos or virtual reality (VR) contents. A novel two-tier streaming framework is proposed to simultaneously address the dynamics in bandwidth variation and user viewing direction changes. The proposed solution formulates the 360-degree video streaming as a dynamic video segment scheduling problem and utilizes prioritized buffer control solution to effectively determine the pre-fetching action and bitrate, based on the current estimated bandwidth, buffer status and view prediction accuracy. The system achieves a significant performance improvement in the delivered video quality compared with benchmark solutions.