Rate-distortion analysis of nonlinear quantisers for MPEG video coders: sigmoidal and unimodal quantiser control functions

摘要

Quantisation is employed in MPEG video coders as a video rate control scheme to regulate the data rate of the compressed video bit stream entering the transmission buffer. For constant bit rate applications, the quantiser has a crucial effect on the video data rate and video quality. It has been a challenging task to optimise the quantiser step size for both bit rate and quality since they are mutually exclusive parameters, as defined by rate-distortion theory. The quantiser step size is generally determined by a linear relationship with respect to the buffer occupancy. Two nonlinear quantiser control functions are investigated, sigmoidal and unimodal, which achieve superior video rate control performance while maintaining similar video quality to the linear one. These two functions are analysed in the framework of rate-distortion theory. Their performance for video rate fluctuation has also been analysed. Encoding results for the two functions are compared to the MPEG2 TM5 evaluation model.