Single-pass constant- and variable-bit-rate MPEG-2 video compression

IBM Journal of Research and Development, Jul 1999 by Mohsenian, N, Rajagopalan, R, Gonzales, C A

For consumer video, a popular mode of encoding operation has been the multipass VBR scheme. For example, high-end DVD mastering can be accomplished with an encoding system which typically uses a three-step procedure. The first step encodes the video source in a CBR mode and gathers a set of predefined statistical features. This information is then used to compute a set of optimized quantization parameters which closely match the source distribution of the video data and would provide a better compressed stream during a second-pass encoding. The last step is a postprocessing task which creates the final DVD format. The intermediate procedure can be carried out several times to produce an ideal video program. This type of application has the advantage that many lookahead parameters are known a priori when the VBR video coding algorithm is implemented. Unfortunately, for applications such as home DVD productions, writable DVDs for PC, digital camcorders, and DVCRs, the availability of a platform capable of analyzing and performing a multipass VBR encoding is unrealistic. Therefore, a VBR scheme is desired which is superior to CBR and can be implemented in one pass. This paper is intended to introduce such an algorithm.

In the first part of the paper we propose a new singlepass CBR encoder algorithm which is tailor-made for realtime compression and can easily be realized with the IBM encoder architecture. We add a further level of complexiity to the CBR encoder and introduce a new real-time singlepass VBR encoder in the second part of the paper. Our VBR scheme employs a causal predictive model to distinguish the "hardness" or "softness" of the incoming video material on the fly and adapts itself accordingly. Moreover, it relies on a perceptual model to improve

the quality of the stressful segments of the stream and produce high-fidelity video. The perceptual model is also responsible for adjustment of the average rate of the stream. The rest of the paper is organized as follows. Section 2 gives some background information for CBR and VBR algorithms. The new CBR encoding technique is presented in Section 3, and the VBR rate-control algorithm is defined in Section 4. Simulation results are given in Section 5, and concluding remarks are provided in Section 6.

2. Background formulation

Again, a guard band gu is used to prevent the decoder buffer from underflowing. After the new target for a picture is set through the conditions given in Equation (22), we must still clip the bits using the adjusted lower

bound L,.adj and the upper bound Un. The updated parameters D,, adi and E,,.adj are used to ensure that the ne) targets are achieved by implementing a macroblock-level rate-control strategy, as defined in (15). Figure 3 displays a graphical representation of how a transformation of a plot of the relative complexity of a picture is used to set a new bit target. It further reflects how VBV buffer compliance is guaranteed and picture bit budgets are satisfied by enforcing the actual bit production to operate within certain limits and cling to the ideal bit allocation.