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

One way to improve the perceptual quality of a CBR stream while maintaining its constant rate from start to finish is to identify "difficult-to-encode" pictures and increase their bit budget accordingly. Conventional approaches to real-time CBR encoding use picture-topicture; correlations in terms of complexity measures to predict the level of difficulty of a picture. In this paper we improve upon this first-order prediction by estimating the encoding difficulty of a picture on the basis of the complexity of all previously encoded pictures. Our singlepass CBR algorithm employs an infinite impulse response (IIR) filter to dynamically determine a nominal value which represents the degree of difficulty of the partially analyzed video stream in real time. We claim that the number of bits consumed by a more (or less) "difficultto-encode" picture can be adjusted by comparing the encoding difficulty of the substream against a local measurement. This method of real-time compression improves the overall quality of the video and maintains a constant rate throughout the stream. Therefore, it can potentially become a key encoding element in cable television (CATV), direct satellite, and terrestrial broadcast arenas as well as mobile and asynchronous transfer mode (ATM) communications.

Since we argued that the video is inherently variable, an even better compressed stream can be created by employing a variable-bit-rate (VBR) encoder algorithm. Applications for such a scheme are plentiful. VBR can be used for networks that employ a dynamic bandwidth, as in ATMs, or it can be exploited as a means of achieving statistical multiplexing for digital broadcast satellites. Other major arenas which can benefit from the use

of a VBR encoder are consumer video and interactive multimedia, where recording of high-quality pictures onto a storage medium is desired. For the aforementioned environments, DVD video movies are created to be played on set-top players, while DVD-RAM drives are used for multimedia productions in computers. Some examples of end-to-end solutions of the above industry segments are shown in Figure 1.

Some attempts at formulating a VBR video coding scheme have appeared in the literature. Most of them propose constraining the quantizer-scale parameter to a user-defined target value for long periods of time [3]. This value is adjusted via network negotiation or monitoring the system buffer to match quality of service (QoS). Although simple intuition suggests that fixing the quantizer sealer would redistribute the amount of bits among GOPs of differing complexity, there are no guarantees of obtaining a constant video quality. Further, in ATM applications an encoded stream is usually queued before a bandwidth is available in the network. This lead time enables pre-encoding tasks to be performed on particular streams subject to the overall channel bandwidth and buffer fullness. Others have investigated

a simple variable-bandwidth estimation model for the number of ATM cells generated through packetization of video sources [4]. In this allocation scheme, a fixed quantizer-scale parameter is again used by the encoder. Authors in [51 have presented a VBR video encoder which takes advantage of the limits of the human observer to improve the perceived quality of the decoded sequence while maint the output bit rate within permitted bounds.