Optimal Allocation of Bits to Frames in Compressed Video

This paper concerns the optimal allocation of bits to individual video frames when compressing a video sequence, where the average number of bits per frame is given as a prior constraint. The optimization algorithm developed here is suitable for offline compression; it requires multiple passes through the video sequence. We hope to extend this work to the case of online (single-pass) compression in a future paper. Our algorithm—which falls under the general heading of rate control mechanisms—was developed in the context of the Strathclyde Compression Transform (SCT) [4, 5]. To date the SCT has employed a rate control that minimizes end-to-end delay rather than maximizing perceptual quality. On the basis that it takes a finite time to transmit each bit in a “real-time” mode of operation, each frame is compressed up to the point where it utilizes the bits available per second (bandwidth) divided by the frames encoded per second (frame rate). While suitable for applications such as video phone where delay is critical, the minimal delay rate control is inappropriate for offline operation where file size rather than transmission delay is an issue. This research was motivated by the hypothesis that a significant improvement in quality ought to be obtainable by varying the allocation of bits to frames in the light of a well-defined optimization procedure, based on the economic principle of equalization of marginal return. For reasons given in section 2.1 below, the SCT is particularly suitable for such a form of rate control. Section 2 presents the relevant theoretical argument, describes the optimization algorithm, and discusses the properties of the algorithm. Section 3 presents experimental results, and section 4 concludes.