Improved UMHexagonS algorithm and architecture for low power H.264 video compression

Abstract

Video has been part of our daily life either for entertainment, work, or communication. The video can be used in form of television, movies, streaming video, video call or even for personal recording. The process of recording and transferring the video data requires a lot of resources such as computational time, storage space and bandwidth (bit rate). This process has become more complex since the demand for better quality and faster video encoding is increasing. The latest video compression standard, H.264, is able to meet this demand but at the cost of increasing computational complexity. This in turn increases the energy consumption of this video compression standard. Motion estimation (ME) is the module that consumes the most of encoding time and computational complexity in video compression. To overcome the increase in computational complexity of ME, H.264 reference software has implemented fast search algorithm known as Unsymmetrical Multi Hexagon-grid Search (UMHexagonS) as the main motion estimation engine. This thesis proposes several improvements for the UMHexagonS in term of algorithms and architectures. The proposed algorithms reduce the computational complexity of the UMHexagonS by reducing the number of search candidate up to 58.54% compared to the conventional UMHexagonS algorithm. It is able to reduce the motion estimation encoding time (MET) up to 28.66% when simulated using H.264 reference software. In addition, the proposed UMHexagonS architectures implement the proposed algorithms efficiently. The proposed architecture is able to reduce the clock cycle up to 87.80% with total energy saving up to 78.79% as compared to the conventional UMHexagonS architecture.