Provendo Eficiência Energética na Codificação de Vídeo usando subtratores aproximados na Predição Interquadros

Abstract

Energy efficiency in video encoding has become increasingly relevant given the widespread use of video in various modern applications and the associated challenge of transmitting and storing large volumes of video data. In this context, the video en coding process is highlighted, with special attention to Interframe Prediction and the Sum of Absolute Differences (SAD), which are fundamental for efficient video com pression. The research presented in this thesis focuses on enhancing the SAD cal culation through the implementation of approximate subtractors, seeking a balance between precision and energy efficiency. This work is part of the panorama of tech nological advancements in video encoders, emphasizing the need for solutions that reduce energy consumption without significantly impacting video quality. The contri butions of this thesis include the development of an innovative approximate subtractor, named Approximate Subtractor (AppS), and the implementation of a hardware system dedicated to interframe prediction in the Versatile Video Coding (VVC) standard. A detailed analysis of approximate subtractors is carried out, including the AppS. The presented research encompasses a comprehensive evaluation of these subtrac tors, analyzing their impact on encoding quality and energy efficiency, through tests in various scenarios and encoding settings. Additionally, the study explores the efficiency of the Test Zone Search (TZS) al gorithm in the context of approximate subtractors, examining how they influence the accuracy of motion vectors and the overall encoding process. In the study on the energy efficiency and performance of the Sum of Absolute Dif ferences (SAD) operations in approximate subtractors, the results varied significantly among the analyzed models. The AppS1 subtractor demonstrated consistency and high efficacy, with constant performance and significant reduction in energy consump tion. In contrast, the AXCS24 recorded an increase in SAD operations and greater energy consumption, highlighting the importance of careful selection of subtractors to optimize both energy efficiency and precision in video encoding, as detailed in the tables of the presented thesis.