Video rate control using a radial basis function estimator for constant bit-rate MPEG coders

Abstract

Compressed video is a source of bursty traffic in communication networks whose data rate needs to be controlled within the available channel capacity, particularly, when it is transmitted via a fixed rate channel. Since the video rate is nonstationary and bursty at large-scene variations in a statistical sense, we propose a feed-forward, estimator-based rate control scheme associated with spatio-temporal activity features (STAF) for MPEG video encoders. This information is used to estimate the video rate of input picture frames. The estimated video rate enables the future buffer occupancy to be calculated and permits the encoder to adapt the quantisation step size to limit the increase or decrease in video rate due to dramatic scene variation. The current and future occupancies are used in a nonlinear quantiser control scheme to determine an appropriate quantisation step size depending on them. The novelty of this technique is that the nonlinear prediction and the nonlinear quantiser control are combined to achieve effective feed-forward video rate control, particularly, for realistic video containing various scene variations. In this paper, we highlight the innovative structure of the scheme and evaluate the performance of rate control algorithms with heuristic, linear and nonlinear rate estimators in the framework of the MPEG2 test model 5 video encoder. The performance measures are the occupancy of a two-frame delay buffer and peak signal-to-noise ratio (PSNR) for video quality.