Computational graceful degradation for video decoding

The implementation of software based video decoders is an advantageous solution over traditional dedicated hardware real-time systems. However, guaranteeing real-time performance, needed when processing video/audio bit-streams, by scheduling processing tasks of variable load and that might exceed the available resources, is an extremely difficult and in many cases an impossible challenge. In this paper we introduce the motivations and ideas of the Computational Graceful Degradation (CGD) approach for the implementation of video decoding. The goal is to provide results of possibly lower visual quality, but respecting the real-time constraints, when the required processing needs exceed the available processing resources. This approach is very attractive since it enables a more efficient usage of the hardware processing power without needing to implement resources for worst case complexity decoding. We show how such technique can be applied to the decoding of compressed video sequences. Simulation results for MPEG-4 and H.263 video compression standards showing some of the interesting achievements and potentialities of CGD approach are also reported. The presented approach can be virtually applied to any video compression standard and processor based platform, although the best performances are achieved when complexity prediction information provided as side information by the video standard is available. This option has been included in the new MPEG-4 standard.