Current implementations of real-time video decoders employhard-decoding without run-time computation regulation. The result isfluctuation in the video decoding time that requires over-specifieddedicated hardware or general-purpose processors to guarantee real-timeperformance. We propose a computation regulation system for MPEG-2 videodecoding on media processors that obviates over-engineering whilemaintaining real-time performance and video quality. Our computationregulation scheme consists of dynamic complexity prediction and control.The complexity prediction algorithm takes implementation issues specificto media processors into account and yields accurate prediction resultsthat correlate with actual measurements to 97%. The predicted complexityis then used as a control signal to adjust the decoding algorithm sothat peaks in computation load are suppressed. Simulation results showthat our system can lower the overall CPU cycle budget by a large factor(25%) with very little degradation in video quality
展开▼
