Many embedded video-based systems require a video codec to reduce the bitrate prior to exchange video information. MPEG High Efficiency Video Coding (HEVC) is the latest, most efficient codec developed by the MPEG group. In the context of HEVC decoding, the optimization of the motion compensation stage is a daunting task. Recently, software implementation studies tackled it leveraging on approximate computing. This paper intends to prove the possibility of adopting coarse-grained hardware reconfiguration to provide dynamic energy management to an HEVC decoder. Runtime coarse-grained adaptation is shown to guarantee energy reduction, in constraint-aware or user-defined situations, while introducing a controllable quality degradation due to approximation.
Runtime Energy versus Quality Tuning in Motion Compensation Filters for HEVC / Palumbo, Francesca; Sau, Carlo; Evangelista, Davide; Meloni, Paolo; Pelcat, Maxime; Raffo, Luigi. - 49:25(2016), pp. 145-152. (Intervento presentato al convegno Programmable Devices and Embedded Systems tenutosi a Brno, CZECH REPUBLIC nel 05-07/10/2016) [10.1016/j.ifacol.2016.12.025].
Runtime Energy versus Quality Tuning in Motion Compensation Filters for HEVC
PALUMBO, Francesca;
2016-01-01
Abstract
Many embedded video-based systems require a video codec to reduce the bitrate prior to exchange video information. MPEG High Efficiency Video Coding (HEVC) is the latest, most efficient codec developed by the MPEG group. In the context of HEVC decoding, the optimization of the motion compensation stage is a daunting task. Recently, software implementation studies tackled it leveraging on approximate computing. This paper intends to prove the possibility of adopting coarse-grained hardware reconfiguration to provide dynamic energy management to an HEVC decoder. Runtime coarse-grained adaptation is shown to guarantee energy reduction, in constraint-aware or user-defined situations, while introducing a controllable quality degradation due to approximation.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
