Dynamic Switching of GOP Configurations in High Efficiency Video Coding (HEVC) using Relational Databases for Multi-objective Optimization

Our current technological era is flooded with smart devices that provide significant computational resources that require optimal video communications solutions. Optimal and dynamic management of video bitrate, quality and energy needs to take into account their inter-dependencies. With emerging network generations providing higher bandwidth rates, there is also a growing need to communicate video with the best quality subject to the availability of resources such as computational power and available bandwidth. Similarly, for accommodating multiple users, there is a need to minimize bitrate requirements while sustaining video quality for reasonable encoding times. This thesis focuses on providing an efficient mechanism for deriving optimal solutions for High Efficiency Video Coding (HEVC) based on dynamic switching of GOP configurations. The approach provides a basic system for multi-objective optimization approach with constraints on power, video quality and bitrate. This is accomplished by utilizing a recently introduced framework known as Dynamically Reconfigurable Architectures for Time-varying Image Constraints (DRASTIC) in HEVC/H.265 encoder with six different GOP configurations to support optimization modes for minimum rate, maximum quality and minimum computational time (minimum energy in constant power configuration) mode of operation. Pareto-optimal GOP configurations are used in implementing the DRASTIC modes. Additionally, this thesis also presents a relational database formulation for supporting multiple devices that are characterized by different screen resolutions and computational resources. This approach is applicable to internet-based video streaming to different devices where the videos have been pre-compressed. Here, the video configuration modes are determined based on the application of database queries applied to relational databases. The database queries are used to retrieve a Pareto-optimal configuration based on real-time user requirements, device, and network constraints