The USC Andrew and Erna Viterbi School of Engineering USC Signal and Image Processing Institute USC Ming Hsieh Department of Electrical Engineering University of Southern California

Technical Report USC-SIPI-308

“Video Bit-Rate Control with Spline Approximated Rate-Distortion Characteristics”

by Liang-Jin Lin

May 1997

Digital video's increased popularity has been driven to a large extent by a flurry of recently proposed international standards (MPEG-1, MPEG-2, H.263, etc.). To transmit video streams over a communication channel, it is required to control the compression in the encoder such that the output data rate can fit to the channel. The rate control scheme is not defined in the standard and thus different strategies can be implemented in each encoder design. The control scheme defined in MPEG Test Model 5 provides a solution with very light computational overhead. However, the results are not always good for any given video sequence and channel rates. It is also difficult to adjust parameters to improve the quality without the help of some trial and error encoding tests. Several rate-distortion (R-D) based techniques have been proposed to solve these problems and generate better and more stable results. However, these approaches are complex because they require the R-D characteristics of the input data to be measured before making quantization assignment decisions.

In this research, we show how the complexity of computing the R-D data can be reduced without reducing too much the performance of the optimization procedure. We propose three stages which provide successive reductions in complexity. In the first stage, we propose an algorithm based on penalty functions and iterative gradient search, which requires less computations than those of trellis-based approaches. The second stage focuses on the approximation of rate and distortion functions. We propose an approximation method based on computing a few R-D points and interpolating the remaining points using spline functions and inter-frame dependency models. In the final stage, we propose a fast algorithm suitable for real-time encoding. The experimental results shows that better and more stable quality can be achieved by our algorithms, especially when the channel rate is low (e.g., CIF format at 192 to 256 kbits per second). All our algorithms and encoding results are compatible to standard MPEG decoder, hence can be played back by any MPEG decoder and get improved quality.

To download the report in PDF format click here: USC-SIPI-308.pdf (22.4Mb)